Tabula Sapiens 2.0 Atlas Maps 1.1 Million Cells Across 28 Tissues
On September 2, the journal Cell, September 2 published an article describing the second version of the Tabula Sapiens atlas, which now includes data from 1.1 million cells across 28 human tissues. The update added nine donors and four tissues, and examined RNA profiles of cells that have stopped dividing.
RNA molecules are working copies of genes, showing which genes a cell is currently reading. Because comparing organs across individuals can mix tissue differences with donor‑specific variation and sample handling, the researchers took multiple organs from four of the nine new donors to isolate tissue effects.
To study stable cell‑division arrest (senescence), they defined a working condition: a cell must contain RNA of CDKN2A (a cell‑cycle‑arrest gene) and lack RNA of MKI67 (a marker of active division). Using this criterion they isolated 48,114 cells from 25 tissues of 21 donors.
Within this
🔗 Read original →
On September 2, the journal Cell, September 2 published an article describing the second version of the Tabula Sapiens atlas, which now includes data from 1.1 million cells across 28 human tissues. The update added nine donors and four tissues, and examined RNA profiles of cells that have stopped dividing.
RNA molecules are working copies of genes, showing which genes a cell is currently reading. Because comparing organs across individuals can mix tissue differences with donor‑specific variation and sample handling, the researchers took multiple organs from four of the nine new donors to isolate tissue effects.
To study stable cell‑division arrest (senescence), they defined a working condition: a cell must contain RNA of CDKN2A (a cell‑cycle‑arrest gene) and lack RNA of MKI67 (a marker of active division). Using this criterion they isolated 48,114 cells from 25 tissues of 21 donors.
Within this
🔗 Read original →
We need to translate Russian news post into natural English, format per rules.
First line: short headline under 90 chars, no markdown, no '#'.
Then blank line, then body paragraphs short (2-3 sentences each), separated by blank lines.
Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks . At most 4-5 per post, never a whole sentence.
Wrap study/journal citations and publication references in single underscores.
We need to preserve facts: Tabula Sapiens 2.0 collected atlas of 1.1 million cells from 28 human tissues. On September 2, article about second version of Tabula Sapiens published in journal Cell. They added nine donors and four tissues, and studied RNA profiles of cells with division arrest. RNA are working copies of genes: they show which genes a cell is currently reading. Inter-organ comparisons easily mix tissue properties with differences between individuals and sample processing. Therefore, for four of the nine new donors they took multiple organs. Comparing tissues from same person allows researchers to more clearly separate tissue contribution from donor-related differences. For analysis of senescence — stable cell division arrest — authors set working condition: cell has RNA copy of gene CDKN2A, associated with cell cycle arrest, and lacks RNA copy of MKI67, indicator of active division. Thus they isolated 48,114 cells from 25 tissues of 21 donors. For each combination of cell type, tissue, and donor, their RNA compared with other cells of same group. In this sample they found 3,792 genes with elevated RNA levels. After CDKN2A, the most common was CDKN2B, also related to cell cycle arrest: it increased in 18 of 30 major cell groups. CXCL8 and MIF increased in five such groups, SERPINE1 — in two. Then authors combined these genes into 17 programs — groups of genes whose RNA levels change together. In epithelial cells forming organ coverings and linings, and fibroblasts creating tissue support, adhesion programs are more noticeable. In immune, epithelial, and stem cells, energy exchange programs and systems that assemble and disassemble proteins changed differently. The working criterion gathered the overall sample, and its RNA programs primarily differed together with cell type.
We need to select key facts for double asterisks: numbers: 1.1 million cells, 28 tissues, September 2, Cell journal, nine donors, four tissues, 48,114 cells, 25 tissues, 21 donors, 3,792 genes, CDKN2A, CDKN2B, CXCL8, MIF, SERPINE1, 18 of 30 major cell groups, five groups, two groups, 17 programs. That's many. We need at most 4-5 double asterisks. So we need to be selective: maybe highlight total cells, date, journal, number of donors/tissues added, number of senescent cells isolated, number of elevated genes, maybe top gene. Choose 4-5.
We need to wrap study/journal citations in single underscores: e.g., Cell, September 2 or Cell? The citation: article published in Cell on September 2. So we can put Cell, September 2.
Now produce headline: maybe "Tabula Sapiens 2.0 Atlas Maps 1.1 Million Cells Across 28 Human Tissues". Count characters: Let's count: "Tabula Sapiens 2.0 Atlas Maps 1.1 Million Cells Across 28 Human Tissues". Count: T(1) a2 b3 u4 l5 a6 space7 S8 a9 p10 i11 e12 n13 s14 space15 1 16 . 17 2 18 . 19 0 20 space21 A22 t23 l24 a25 s26 space27 M28 a29 p30 s31 space32 1 33 . 34 1 35 space36 M37 i38 l39 l40 i41 o42 n43 space44 C45 e46 l47 l48 s49 space50 A51 c52 r53 o54 s55 s56 space57 2 58 8 59 space60 H61 u62 m63 a64 n65 space66 T67 i68 s69 s70 u71 e72 s73. That's 73 characters. Under 90, good.
Now blank line.
Now body paragraphs: maybe 3-4 paragraphs each 2-3 sentences.
Paragraph 1: Introduce publication and dataset.…
🔗 Read original →
First line: short headline under 90 chars, no markdown, no '#'.
Then blank line, then body paragraphs short (2-3 sentences each), separated by blank lines.
Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks . At most 4-5 per post, never a whole sentence.
Wrap study/journal citations and publication references in single underscores.
We need to preserve facts: Tabula Sapiens 2.0 collected atlas of 1.1 million cells from 28 human tissues. On September 2, article about second version of Tabula Sapiens published in journal Cell. They added nine donors and four tissues, and studied RNA profiles of cells with division arrest. RNA are working copies of genes: they show which genes a cell is currently reading. Inter-organ comparisons easily mix tissue properties with differences between individuals and sample processing. Therefore, for four of the nine new donors they took multiple organs. Comparing tissues from same person allows researchers to more clearly separate tissue contribution from donor-related differences. For analysis of senescence — stable cell division arrest — authors set working condition: cell has RNA copy of gene CDKN2A, associated with cell cycle arrest, and lacks RNA copy of MKI67, indicator of active division. Thus they isolated 48,114 cells from 25 tissues of 21 donors. For each combination of cell type, tissue, and donor, their RNA compared with other cells of same group. In this sample they found 3,792 genes with elevated RNA levels. After CDKN2A, the most common was CDKN2B, also related to cell cycle arrest: it increased in 18 of 30 major cell groups. CXCL8 and MIF increased in five such groups, SERPINE1 — in two. Then authors combined these genes into 17 programs — groups of genes whose RNA levels change together. In epithelial cells forming organ coverings and linings, and fibroblasts creating tissue support, adhesion programs are more noticeable. In immune, epithelial, and stem cells, energy exchange programs and systems that assemble and disassemble proteins changed differently. The working criterion gathered the overall sample, and its RNA programs primarily differed together with cell type.
We need to select key facts for double asterisks: numbers: 1.1 million cells, 28 tissues, September 2, Cell journal, nine donors, four tissues, 48,114 cells, 25 tissues, 21 donors, 3,792 genes, CDKN2A, CDKN2B, CXCL8, MIF, SERPINE1, 18 of 30 major cell groups, five groups, two groups, 17 programs. That's many. We need at most 4-5 double asterisks. So we need to be selective: maybe highlight total cells, date, journal, number of donors/tissues added, number of senescent cells isolated, number of elevated genes, maybe top gene. Choose 4-5.
We need to wrap study/journal citations in single underscores: e.g., Cell, September 2 or Cell? The citation: article published in Cell on September 2. So we can put Cell, September 2.
Now produce headline: maybe "Tabula Sapiens 2.0 Atlas Maps 1.1 Million Cells Across 28 Human Tissues". Count characters: Let's count: "Tabula Sapiens 2.0 Atlas Maps 1.1 Million Cells Across 28 Human Tissues". Count: T(1) a2 b3 u4 l5 a6 space7 S8 a9 p10 i11 e12 n13 s14 space15 1 16 . 17 2 18 . 19 0 20 space21 A22 t23 l24 a25 s26 space27 M28 a29 p30 s31 space32 1 33 . 34 1 35 space36 M37 i38 l39 l40 i41 o42 n43 space44 C45 e46 l47 l48 s49 space50 A51 c52 r53 o54 s55 s56 space57 2 58 8 59 space60 H61 u62 m63 a64 n65 space66 T67 i68 s69 s70 u71 e72 s73. That's 73 characters. Under 90, good.
Now blank line.
Now body paragraphs: maybe 3-4 paragraphs each 2-3 sentences.
Paragraph 1: Introduce publication and dataset.…
🔗 Read original →
Bacterial Patterns Shift with Moving Frog Embryon and Xenobot, Machine Model Distinguishes Conditions
On September 3, a preprint reported that motile *Bacillus subtilis* bacteria altered their pattern in fluid near a frog embryo and a xenobot — a body assembled from embryonic cells. When the embryo was moved, the bacterial accumulation zone moved with it. The bacterial pattern allowed a computer model to distinguish three conditions: embryo, xenobot, and bacteria‑only culture.
Researchers tested whether the bacterial spatial pattern contains information about a neighboring living system, comparing the frog embryo and xenobot (both share the same species and genome, but the xenobot is assembled separately and develops differently). In bacteria‑only culture, motile cells self‑organized into a branching pattern; a non‑motile strain and fluorescent microparticles hardly produced this pattern. Near the xenobot, motile bacteria formed a glowing halo whose average area reached 9.4 mm² after one hour, while in the two control groups it remained at hundredths of a square millimeter. This linked the halo to active bacterial behavior.
Near the living embryo the halo grew; near a heat‑inactivated embryo the early accumulation then receded. When the embryo was translocated, the bacterial cluster disappeared from the original site and appeared at the new location, showing that the bacterial pattern changes with the target’s position and state.
Researchers then examined the ionic environment. Raising potassium chloride concentration from 1.5 to 200 mmol/L increased the halo area to 23.0 mm² by the 11th hour, up from 0.67 mm² at baseline. Bacteria lacking the YugO potassium‑channel protein showed slower early accumulation, indicating that the ionic milieu modulates the early bacterial response to the living target.
To force the model to rely on the bacterial pattern, the target’s silhouette was masked in the frames. From the remaining bacterial distribution on deposited rollers the model distinguished the three conditions with 61.13% accuracy, compared with a baseline of 33%. The authors conclude that “the physiological state of one collective is partially recorded in the shape of another,” showing that bacterial positioning serves as a measurable readout of the state and position of a neighboring multicellular target.
🔗 Read original →
On September 3, a preprint reported that motile *Bacillus subtilis* bacteria altered their pattern in fluid near a frog embryo and a xenobot — a body assembled from embryonic cells. When the embryo was moved, the bacterial accumulation zone moved with it. The bacterial pattern allowed a computer model to distinguish three conditions: embryo, xenobot, and bacteria‑only culture.
Researchers tested whether the bacterial spatial pattern contains information about a neighboring living system, comparing the frog embryo and xenobot (both share the same species and genome, but the xenobot is assembled separately and develops differently). In bacteria‑only culture, motile cells self‑organized into a branching pattern; a non‑motile strain and fluorescent microparticles hardly produced this pattern. Near the xenobot, motile bacteria formed a glowing halo whose average area reached 9.4 mm² after one hour, while in the two control groups it remained at hundredths of a square millimeter. This linked the halo to active bacterial behavior.
Near the living embryo the halo grew; near a heat‑inactivated embryo the early accumulation then receded. When the embryo was translocated, the bacterial cluster disappeared from the original site and appeared at the new location, showing that the bacterial pattern changes with the target’s position and state.
Researchers then examined the ionic environment. Raising potassium chloride concentration from 1.5 to 200 mmol/L increased the halo area to 23.0 mm² by the 11th hour, up from 0.67 mm² at baseline. Bacteria lacking the YugO potassium‑channel protein showed slower early accumulation, indicating that the ionic milieu modulates the early bacterial response to the living target.
To force the model to rely on the bacterial pattern, the target’s silhouette was masked in the frames. From the remaining bacterial distribution on deposited rollers the model distinguished the three conditions with 61.13% accuracy, compared with a baseline of 33%. The authors conclude that “the physiological state of one collective is partially recorded in the shape of another,” showing that bacterial positioning serves as a measurable readout of the state and position of a neighboring multicellular target.
🔗 Read original →
bioRxiv
Living multicellular systems induce decodable spatial patterns in bacterial collectives
Living systems continuously modify their environments through chemical, mechanical, metabolic and bioelectrical activity. Whether a presence of a multicellular system can be encoded into the emergent spatial organization of another living collective in a…
We need to translate Russian news post into natural English, format per rules.
First line: short headline under 90 chars, no markdown. Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.
Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks. At most 4-5 per post, never a whole sentence.
Wrap study/journal citations and publication references in single underscores (e.g. Nature Aging, July 2026).
We need to preserve all facts, numbers, names, citations exactly. No added commentary.
We need to identify important facts: numbers like 186, 70 million, 4680, 600, 246, 40 fixed additions, 77 rational coefficients, >1/50000 margin, August 2025? Actually date: "30 августа OpenAI опубликовала работу" => August 30 (year unspecified, likely 2025?). The original text: "OpenAI опубликовала результат GPT-6 Astra: бесконечно многие соседние простые различаются не более чем на 186 30 августа OpenAI опубликовала работу о коротких промежутках между простыми." So date: August 30 (maybe 2024?). We'll keep as "August 30". Also mention "2013 year Ethan Zhang", "Polymath projects", "James Maynard", "next Polymath project", "OpenAI work reduces bound to 186". Also mention "Axiom Math formalized previous bound 246 in August". Also mention "Lean 4 code", "numerical certificate", "Selberg sieve", etc.
We need to wrap key numbers in double asterisks: maybe 186, 70 million, 4680, 600, 246, 40, 77, >1/50000. That's many; we need at most 4-5. So we must be selective: choose the most important: the bound 186, the previous bound 246, the improvement from 70 million to 186 maybe mention 70 million as historical. But we can only have 4-5 total. Let's pick: 186, 246, 70 million, 40 (fixed additions), maybe 77 rational coefficients. That's 5.
We must not wrap whole sentences; just wrap the numbers.
Citations: we have references to work by OpenAI, Polymath, James Maynard, Axiom Math formalization, Lean checking. We need to wrap study/journal citations and publication references in single underscores. There's no explicit journal name; maybe we treat the work as a paper? Could use OpenAI work, August 30? But that's not a journal. The instruction: "Wrap study/journal citations and publication references in single underscores (e.g. Nature Aging, July 2026)." So we need to identify any such citations. The text mentions "OpenAI опубликовала работу", "В августе Axiom Math формализовала прежнюю границу 246 ; там Lean проверял цепочку уже известной теоремы." Also "В новом репозитории эта цепочка начинается с трёх явно названных входов". Not exactly journal citations. Could treat "OpenAI work, August 30" as a publication reference. Also "Axiom Math, August". We'll wrap those in single underscores.
We must not add source link.
Now produce headline: maybe "OpenAI’s GPT-6 Astra proves infinitely many prime gaps ≤ 186". Count characters: Let's count: "OpenAI’s GPT-6 Astra proves infinitely many prime gaps ≤ 186". That's about 58 characters. Good.
Now body: need short paragraphs 2-3 sentences each.
We'll produce maybe 4 paragraphs.
Paragraph 1: Summarize the result: On August 30, OpenAI released a result from GPT-6 Astra showing that infinitely many consecutive primes differ by at most 186, i.e., lim inf (p_{n+1} - p_n) ≤ 186. Include citation OpenAI work, August 30.…
🔗 Read original →
First line: short headline under 90 chars, no markdown. Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.
Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks. At most 4-5 per post, never a whole sentence.
Wrap study/journal citations and publication references in single underscores (e.g. Nature Aging, July 2026).
We need to preserve all facts, numbers, names, citations exactly. No added commentary.
We need to identify important facts: numbers like 186, 70 million, 4680, 600, 246, 40 fixed additions, 77 rational coefficients, >1/50000 margin, August 2025? Actually date: "30 августа OpenAI опубликовала работу" => August 30 (year unspecified, likely 2025?). The original text: "OpenAI опубликовала результат GPT-6 Astra: бесконечно многие соседние простые различаются не более чем на 186 30 августа OpenAI опубликовала работу о коротких промежутках между простыми." So date: August 30 (maybe 2024?). We'll keep as "August 30". Also mention "2013 year Ethan Zhang", "Polymath projects", "James Maynard", "next Polymath project", "OpenAI work reduces bound to 186". Also mention "Axiom Math formalized previous bound 246 in August". Also mention "Lean 4 code", "numerical certificate", "Selberg sieve", etc.
We need to wrap key numbers in double asterisks: maybe 186, 70 million, 4680, 600, 246, 40, 77, >1/50000. That's many; we need at most 4-5. So we must be selective: choose the most important: the bound 186, the previous bound 246, the improvement from 70 million to 186 maybe mention 70 million as historical. But we can only have 4-5 total. Let's pick: 186, 246, 70 million, 40 (fixed additions), maybe 77 rational coefficients. That's 5.
We must not wrap whole sentences; just wrap the numbers.
Citations: we have references to work by OpenAI, Polymath, James Maynard, Axiom Math formalization, Lean checking. We need to wrap study/journal citations and publication references in single underscores. There's no explicit journal name; maybe we treat the work as a paper? Could use OpenAI work, August 30? But that's not a journal. The instruction: "Wrap study/journal citations and publication references in single underscores (e.g. Nature Aging, July 2026)." So we need to identify any such citations. The text mentions "OpenAI опубликовала работу", "В августе Axiom Math формализовала прежнюю границу 246 ; там Lean проверял цепочку уже известной теоремы." Also "В новом репозитории эта цепочка начинается с трёх явно названных входов". Not exactly journal citations. Could treat "OpenAI work, August 30" as a publication reference. Also "Axiom Math, August". We'll wrap those in single underscores.
We must not add source link.
Now produce headline: maybe "OpenAI’s GPT-6 Astra proves infinitely many prime gaps ≤ 186". Count characters: Let's count: "OpenAI’s GPT-6 Astra proves infinitely many prime gaps ≤ 186". That's about 58 characters. Good.
Now body: need short paragraphs 2-3 sentences each.
We'll produce maybe 4 paragraphs.
Paragraph 1: Summarize the result: On August 30, OpenAI released a result from GPT-6 Astra showing that infinitely many consecutive primes differ by at most 186, i.e., lim inf (p_{n+1} - p_n) ≤ 186. Include citation OpenAI work, August 30.…
🔗 Read original →
We need to translate Russian news post into English, format per rules.
First line: short headline under 90 chars, no markdown, not prefixed with '#'.
Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.
Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks. At most 4-5 per post, never a whole sentence.
Wrap study/journal citations and publication references in single underscores (e.g. Nature Aging, July 2026).
We need to preserve all facts, numbers, names, citations exactly. No added commentary.
We need to identify important facts: numbers: 186, 30 августа (August 30), 2013, 70 миллионов (70 million), 4680, 600, 246, 40 фиксированных добавок (40 fixed offsets), 0 and 186, 77 рациональных коэффициентов (77 rational coefficients), запас больше 1/50000 (margin greater than 1/50000), August Axiom Math formalized previous bound 246, etc.
We need to select at most 4-5 double asterisk facts. Choose the most important: 186, August 30, 2013, 70 million, 246 maybe. But we need at most 4-5. Let's pick: 186, August 30, 2013, 70 million, 246. That's 5.
We need to ensure we don't wrap whole sentences; just wrap the numbers/dates.
We need to wrap study/journal citations in single underscores. There's mention of "OpenAI опубликовала работу о коротких промежутках между простыми". Not a journal. There's "Axiom Math". Possibly a journal? It says "В августе Axiom Math формализовала прежнюю границу 246". So we can wrap Axiom Math as a citation? It's a publication reference. Also maybe "OpenAI" is not a journal. We'll wrap Axiom Math in single underscores.
Also maybe "Lean 4" is a tool, not citation. Not needed.
We need to preserve all facts exactly. So we need to translate the content.
Let's draft headline: "OpenAI’s GPT-6 Astra proves infinitely many prime gaps ≤ 186". Count characters: Let's count: "OpenAI’s GPT-6 Astra proves infinitely many prime gaps ≤ 186". That's about 58 characters. Good.
Now body paragraphs.
We need to split into short paragraphs 2-3 sentences each.
We'll produce maybe 4 paragraphs.
Paragraph 1: On August 30, OpenAI released a result from GPT-6 Astra showing that infinitely many adjacent primes differ by no more than 186. The proof was produced by GPT-6 Astra, accompanied by a numerical certificate and Lean 4 code that checks the logical steps.
Paragraph 2: The number 186 means that there are infinitely many pairs of consecutive primes with a gap at most 186; formally, lim inf (p_{n+1} − p_n) ≤ 186. In 2013, Yitang Zhang established an upper bound of 70 million; subsequent Polymath projects reduced it to 4680, James Maynard to 600, and another Polymath effort to 246. OpenAI’s work improves this bound to 186.
Paragraph 3: The authors select 40 fixed shifts added to a base integer n, all lying between 0 and 186, and prove that for infinitely many n at least two of the shifted numbers are prime, giving a gap ≤ 186. The proof uses a multidimensional Selberg sieve, assigning numerical weights to such sets and estimating prime frequency. A new technique factors common divisors into three parts of suitable size, allowing a broader weight set and the needed inequality for the 40 numbers.…
🔗 Read original →
First line: short headline under 90 chars, no markdown, not prefixed with '#'.
Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.
Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks. At most 4-5 per post, never a whole sentence.
Wrap study/journal citations and publication references in single underscores (e.g. Nature Aging, July 2026).
We need to preserve all facts, numbers, names, citations exactly. No added commentary.
We need to identify important facts: numbers: 186, 30 августа (August 30), 2013, 70 миллионов (70 million), 4680, 600, 246, 40 фиксированных добавок (40 fixed offsets), 0 and 186, 77 рациональных коэффициентов (77 rational coefficients), запас больше 1/50000 (margin greater than 1/50000), August Axiom Math formalized previous bound 246, etc.
We need to select at most 4-5 double asterisk facts. Choose the most important: 186, August 30, 2013, 70 million, 246 maybe. But we need at most 4-5. Let's pick: 186, August 30, 2013, 70 million, 246. That's 5.
We need to ensure we don't wrap whole sentences; just wrap the numbers/dates.
We need to wrap study/journal citations in single underscores. There's mention of "OpenAI опубликовала работу о коротких промежутках между простыми". Not a journal. There's "Axiom Math". Possibly a journal? It says "В августе Axiom Math формализовала прежнюю границу 246". So we can wrap Axiom Math as a citation? It's a publication reference. Also maybe "OpenAI" is not a journal. We'll wrap Axiom Math in single underscores.
Also maybe "Lean 4" is a tool, not citation. Not needed.
We need to preserve all facts exactly. So we need to translate the content.
Let's draft headline: "OpenAI’s GPT-6 Astra proves infinitely many prime gaps ≤ 186". Count characters: Let's count: "OpenAI’s GPT-6 Astra proves infinitely many prime gaps ≤ 186". That's about 58 characters. Good.
Now body paragraphs.
We need to split into short paragraphs 2-3 sentences each.
We'll produce maybe 4 paragraphs.
Paragraph 1: On August 30, OpenAI released a result from GPT-6 Astra showing that infinitely many adjacent primes differ by no more than 186. The proof was produced by GPT-6 Astra, accompanied by a numerical certificate and Lean 4 code that checks the logical steps.
Paragraph 2: The number 186 means that there are infinitely many pairs of consecutive primes with a gap at most 186; formally, lim inf (p_{n+1} − p_n) ≤ 186. In 2013, Yitang Zhang established an upper bound of 70 million; subsequent Polymath projects reduced it to 4680, James Maynard to 600, and another Polymath effort to 246. OpenAI’s work improves this bound to 186.
Paragraph 3: The authors select 40 fixed shifts added to a base integer n, all lying between 0 and 186, and prove that for infinitely many n at least two of the shifted numbers are prime, giving a gap ≤ 186. The proof uses a multidimensional Selberg sieve, assigning numerical weights to such sets and estimating prime frequency. A new technique factors common divisors into three parts of suitable size, allowing a broader weight set and the needed inequality for the 40 numbers.…
🔗 Read original →
We need to translate Russian news post into English, format as per rules.
First line: short headline under 90 chars, no markdown, no '#'.
Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.
Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks . At most 4-5 per post, never a whole sentence.
Wrap study/journal citations and publication references in single underscores _ (e.g., Nature Aging, July 2026).
We need to preserve all facts, numbers, names, citations exactly.
Let's extract facts:
- Date: 29 августа (August 29) in Molecular Neurodegeneration (journal). So citation: Molecular Neurodegeneration, August 29? Actually they said "29 августа в Molecular Neurodegeneration вышла работа Эмили Том и соавторов". So we can put Molecular Neurodegeneration, August 29 (year? Not given; assume 2026? Not given. We'll just keep date as given: August 29. Could also include year if known but not in text. We'll keep as is.
- Authors: Emily Tom and coauthors.
- Study: pigment epithelium of retina.
- They tracked pathway from age-related lipid change to membrane repair and tissue remodeling beneath this layer.
- Pigment epithelium supplies photoreceptors and digests their spent parts.
- Between it and blood vessels lies Bruch's membrane; with age extracellular material accumulates at this boundary.
- In donor epithelium samples and retina of old mice, authors first saw age signature: genes related to membranes changed, and reserve of long-chain polyunsaturated fatty acids.
- Then they checked ELOVL2 — enzyme that elongates such fatty acids.
- When its activity reduced in cells or disrupted in mice, these lipids decreased, and ceramides increased.
- Membrane became more ordered: its lipids packed tighter, calcium level rose in cells.
- Simultaneously lysosomes accumulated at cell edge and fused with outer membrane, releasing part of contents outward.
- Lysosomes normally digest cellular material, but here authors consider their exit to surface as a way to patch damaged boundary.
- In cell layer where upper and lower sides can be measured separately, more acid sphingomyelinase was secreted downward, toward Bruch's membrane.
- This enzyme converts one membrane lipid into ceramides; in mutant mice at same boundary APOE protein accumulated, part of deposits.
- So authors link prolonged lysosomal repair with remodeling of extracellular space beneath pigment epithelium.
- Authors checked this chain in reverse, adding 24:5n-3 — direct product of ELOVL2.
- In 2025 work same team showed in old mice that its administration improved visual indicators and decreased deposits under pigment epithelium.
- In new work addition of 24:5n-3 to cells weakened signs of membrane stress and returned lysosomes from cell edge; after intravitreal injection in mice, ceramide accumulation decreased and pigment epithelium structure partially restored.
- By authors' model, deficiency of ELOVL2 products changes membrane properties, while prolonged lysosomal work maintains its integrity and simultaneously changes tissue beneath pigment epithelium.
We need to wrap key numbers, percentages, drug/company/gene names, dates in double asterisks. Let's identify:
- Date: August 29 (maybe year unknown). Could wrap August 29.
- Gene/enzyme: ELOVL2 (gene name). Also maybe 24:5n-3 (fatty acid). Also APOE (protein). Also maybe acid sphingomyelinase (enzyme). But we need at most 4-5 double asterisks total. Choose most important: ELOVL2, 24:5n-3, APOE, August 29 (date). That's four.
We could also include Molecular Neurodegeneration as journal name? That's a citation, should be wrapped in single underscores, not double asterisks. So we keep that as Molecular Neurodegeneration.…
🔗 Read original →
First line: short headline under 90 chars, no markdown, no '#'.
Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.
Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks . At most 4-5 per post, never a whole sentence.
Wrap study/journal citations and publication references in single underscores _ (e.g., Nature Aging, July 2026).
We need to preserve all facts, numbers, names, citations exactly.
Let's extract facts:
- Date: 29 августа (August 29) in Molecular Neurodegeneration (journal). So citation: Molecular Neurodegeneration, August 29? Actually they said "29 августа в Molecular Neurodegeneration вышла работа Эмили Том и соавторов". So we can put Molecular Neurodegeneration, August 29 (year? Not given; assume 2026? Not given. We'll just keep date as given: August 29. Could also include year if known but not in text. We'll keep as is.
- Authors: Emily Tom and coauthors.
- Study: pigment epithelium of retina.
- They tracked pathway from age-related lipid change to membrane repair and tissue remodeling beneath this layer.
- Pigment epithelium supplies photoreceptors and digests their spent parts.
- Between it and blood vessels lies Bruch's membrane; with age extracellular material accumulates at this boundary.
- In donor epithelium samples and retina of old mice, authors first saw age signature: genes related to membranes changed, and reserve of long-chain polyunsaturated fatty acids.
- Then they checked ELOVL2 — enzyme that elongates such fatty acids.
- When its activity reduced in cells or disrupted in mice, these lipids decreased, and ceramides increased.
- Membrane became more ordered: its lipids packed tighter, calcium level rose in cells.
- Simultaneously lysosomes accumulated at cell edge and fused with outer membrane, releasing part of contents outward.
- Lysosomes normally digest cellular material, but here authors consider their exit to surface as a way to patch damaged boundary.
- In cell layer where upper and lower sides can be measured separately, more acid sphingomyelinase was secreted downward, toward Bruch's membrane.
- This enzyme converts one membrane lipid into ceramides; in mutant mice at same boundary APOE protein accumulated, part of deposits.
- So authors link prolonged lysosomal repair with remodeling of extracellular space beneath pigment epithelium.
- Authors checked this chain in reverse, adding 24:5n-3 — direct product of ELOVL2.
- In 2025 work same team showed in old mice that its administration improved visual indicators and decreased deposits under pigment epithelium.
- In new work addition of 24:5n-3 to cells weakened signs of membrane stress and returned lysosomes from cell edge; after intravitreal injection in mice, ceramide accumulation decreased and pigment epithelium structure partially restored.
- By authors' model, deficiency of ELOVL2 products changes membrane properties, while prolonged lysosomal work maintains its integrity and simultaneously changes tissue beneath pigment epithelium.
We need to wrap key numbers, percentages, drug/company/gene names, dates in double asterisks. Let's identify:
- Date: August 29 (maybe year unknown). Could wrap August 29.
- Gene/enzyme: ELOVL2 (gene name). Also maybe 24:5n-3 (fatty acid). Also APOE (protein). Also maybe acid sphingomyelinase (enzyme). But we need at most 4-5 double asterisks total. Choose most important: ELOVL2, 24:5n-3, APOE, August 29 (date). That's four.
We could also include Molecular Neurodegeneration as journal name? That's a citation, should be wrapped in single underscores, not double asterisks. So we keep that as Molecular Neurodegeneration.…
🔗 Read original →
PubMed Central (PMC)
The lipid elongation enzyme ELOVL2 is a molecular regulator of aging in the retina
Methylation of the regulatory region of the elongation of very‐long‐chain fatty acids‐like 2 (ELOVL2) gene, an enzyme involved in elongation of long‐chain polyunsaturated fatty acids, is one of the most robust biomarkers of human age, but the ...
Essay on LessWrong Suggests Partial Brain Emulation May Boost AI Before Full Copy
On September 5, author TsviBT published an essay on LessWrong about whole brain emulation — a computer model that could replicate a person over long periods. He proposes measuring progress toward such a model by the intermediate data, methods, and simulations that appear earlier and who might benefit from them.
Partial emulations, which reproduce individual brain abilities but not a full personality, are easier to achieve than a complete copy. TsviBT calls the gap between useful fragments and a digital personality the “bad knee,” noting that these fragments could accelerate AI development before a full emulation exists, potentially outweighing its future benefits.
The discussion under the essay highlights that risk depends on the research route, contrasting models trained on brain activity recordings with connectomics — maps of neurons and their connections. The former learns to reproduce brain function directly; the latter describes the brain’s wiring. A Technical review 2025 explains this difference: connectivity maps show wiring, while functional models require activity data, suggesting training on organisms with both data types before scaling to larger mammalian brains.
TsviBT recommends evaluating each emulation program by the intermediate models, data, and methods it makes available prior to achieving full emulation.
🔗 Read original →
On September 5, author TsviBT published an essay on LessWrong about whole brain emulation — a computer model that could replicate a person over long periods. He proposes measuring progress toward such a model by the intermediate data, methods, and simulations that appear earlier and who might benefit from them.
Partial emulations, which reproduce individual brain abilities but not a full personality, are easier to achieve than a complete copy. TsviBT calls the gap between useful fragments and a digital personality the “bad knee,” noting that these fragments could accelerate AI development before a full emulation exists, potentially outweighing its future benefits.
The discussion under the essay highlights that risk depends on the research route, contrasting models trained on brain activity recordings with connectomics — maps of neurons and their connections. The former learns to reproduce brain function directly; the latter describes the brain’s wiring. A Technical review 2025 explains this difference: connectivity maps show wiring, while functional models require activity data, suggesting training on organisms with both data types before scaling to larger mammalian brains.
TsviBT recommends evaluating each emulation program by the intermediate models, data, and methods it makes available prior to achieving full emulation.
🔗 Read original →
Three Safety Checks Proposed for Self‑Driving Labs Before Real Experiments
An essay posted on LessWrong on September 4 describes a laboratory where a model selects the next experimental step and robotic equipment carries it out. The author argues that the entire chain turning a calculation into instrument commands must be examined for safety.
The essay separates the model’s capabilities from its authority. A model can detect danger and devise a plan, but access to materials, instruments, and settings decides which actions the plan will actually trigger in the physical lab. A safety check must consider both what the model can do and what the surrounding system permits.
Three sequential checks are proposed. First, test the model on recognizing hazardous directions and obeying given bans. Second, run the full experiment plan through a virtual model that incorporates instrument limits, incomplete data, and failures to see whether it remains safe. Third, perform a low‑risk task on real hardware to compare expected safety with observed results and refine the calculation before tackling more complex experiments.
In the LabShield benchmark suite, models scored 32.0 percentage points lower on average across 164 tasks in professional scenarios requiring independent reasoning than in multiple‑choice questions. This gap shows why the whole plan — including the model’s ability to give a safe answer — should be vetted before physical work.
After the initial three stages, the author recommends ongoing evaluation. Models, equipment, and software are updated, and new failure modes emerge in the lab; therefore the agent’s authority, instrument status, and ability to abort a procedure must be reassessed regularly.
🔗 Read original →
An essay posted on LessWrong on September 4 describes a laboratory where a model selects the next experimental step and robotic equipment carries it out. The author argues that the entire chain turning a calculation into instrument commands must be examined for safety.
The essay separates the model’s capabilities from its authority. A model can detect danger and devise a plan, but access to materials, instruments, and settings decides which actions the plan will actually trigger in the physical lab. A safety check must consider both what the model can do and what the surrounding system permits.
Three sequential checks are proposed. First, test the model on recognizing hazardous directions and obeying given bans. Second, run the full experiment plan through a virtual model that incorporates instrument limits, incomplete data, and failures to see whether it remains safe. Third, perform a low‑risk task on real hardware to compare expected safety with observed results and refine the calculation before tackling more complex experiments.
In the LabShield benchmark suite, models scored 32.0 percentage points lower on average across 164 tasks in professional scenarios requiring independent reasoning than in multiple‑choice questions. This gap shows why the whole plan — including the model’s ability to give a safe answer — should be vetted before physical work.
After the initial three stages, the author recommends ongoing evaluation. Models, equipment, and software are updated, and new failure modes emerge in the lab; therefore the agent’s authority, instrument status, and ability to abort a procedure must be reassessed regularly.
🔗 Read original →
Nature
Risks of AI scientists: prioritizing safeguarding over autonomy
Nature Communications - AI scientists powered by large language models and AI agents present both opportunities and risks in automatic scientific discovery. Here, the authors examine the...
Hippo–IGF2 pathway controls liver regeneration and tumor growth in mice
On September
🔗 Read original →
On September
🔗 Read original →
Bone marrow vascular endothelial RANK signaling drives age‑related inflammation in mice
On September 3, 2026 Yasuhiro Kobayashi’s group posted a preprint showing how age‑related inflammation begins in mouse bone marrow. They traced a cascade from the vascular‑endothelial protein RANK to the cytokine IL‑1β, which propagates the inflammatory signal. In this model IL‑1β amplified senescence of marrow stromal cells and shifted blood‑cell production toward myeloid lineages.
The team started with the RANKL–RANK–OPG system, where RANKL activates RANK and OPG sequesters RANKL to dampen signaling. In eight‑week mice lacking OPG they observed more senescence‑marker‑positive cells, excess neutrophils and fewer lymphocytes — a phenotype matching that of 32‑week control mice. Treatment with an anti‑RANKL antibody reduced senescence‑marker‑positive cells and alleviated the myeloid skew.
To pinpoint the site of action, the authors referred to their 2025 work linking RANK/RANKL to marrow‑vascular permeability and immune‑cell transmigration. Fluorescently labeled RANKL in 32‑week mice bound chiefly to the endothelium, and deleting RANK only in endothelial cells lowered senescence‑marker‑positive cells and the myeloid bias, while promoting new bone formation after marrow ablation
🔗 Read original →
On September 3, 2026 Yasuhiro Kobayashi’s group posted a preprint showing how age‑related inflammation begins in mouse bone marrow. They traced a cascade from the vascular‑endothelial protein RANK to the cytokine IL‑1β, which propagates the inflammatory signal. In this model IL‑1β amplified senescence of marrow stromal cells and shifted blood‑cell production toward myeloid lineages.
The team started with the RANKL–RANK–OPG system, where RANKL activates RANK and OPG sequesters RANKL to dampen signaling. In eight‑week mice lacking OPG they observed more senescence‑marker‑positive cells, excess neutrophils and fewer lymphocytes — a phenotype matching that of 32‑week control mice. Treatment with an anti‑RANKL antibody reduced senescence‑marker‑positive cells and alleviated the myeloid skew.
To pinpoint the site of action, the authors referred to their 2025 work linking RANK/RANKL to marrow‑vascular permeability and immune‑cell transmigration. Fluorescently labeled RANKL in 32‑week mice bound chiefly to the endothelium, and deleting RANK only in endothelial cells lowered senescence‑marker‑positive cells and the myeloid bias, while promoting new bone formation after marrow ablation
🔗 Read original →
PubMed Central (PMC)
The RANK/RANKL axis controls vascular dynamics in the bone marrow
Receptor activator of nuclear factor kappa B ligand (RANKL) is a multifunctional cytokine and has been attracting high levels of interest in broader fields of biomedical sciences. In this study, we showed that RANKL from C–X–C motif chemokine ligand ...
We need to translate Russian news post into natural English, format per rules.
First line: short headline under 90 chars, no markdown, no '#'.
Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.
Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks. At most 4-5 per post, never a whole sentence.
Wrap study/journal citations and publication references in single underscores (e.g., Nature Aging, July 2026).
We need to preserve facts, numbers, names, citations exactly.
Let's extract facts:
- Study: September 1 in The FASEB Journal (date and journal). So citation: The FASEB Journal, September 1? Actually they said "1 сентября в статье The FASEB Journal". So we need to wrap the citation in single underscores: The FASEB Journal, September 1.
- Researchers compared three light-dark cycles in female mice of one lab strain, fed high-fat diet.
- Each group had eight animals; varied duration of light-dark alternation.
- Mice have internal clocks: in constant darkness activity repeats with its own period called tau.
- Early in experiment tau of mice on high-fat diet was longer than that of mice on low-fat diet, before obesity appeared.
- This observation prompted question: would obesity timing be changed by external light schedule aligned with internal rhythm.
- For testing on high-fat diet they used: usual 24-hour day, cycle about 23.64 hours — close to tau — and cycle 23.28 hours.
- Light and dark parts of each cycle occupied equal time.
- At 24-hour regimen obesity recorded at 4.0 ± 1.1 weeks.
- At two shorter cycles — at 12.1 ± 1.6 and 11.3 ± 1.6 weeks.
- Obesity occurred in all three groups, and shorter days shifted its onset by about eight weeks.
- Accumulated energy intake between light regimens was similar.
- Mice with short cycles maintained higher locomotor activity from second week; body mass differences appeared later.
- Authors link this sequence to higher energy expenditure: timing difference of obesity arose on same high-fat diet and without difference in accumulated energy intake.
- After the shortest regimen, 23.28 hours, internal period in darkness also became shorter.
- After cycle close to tau, it remained at level of the cycle period and constant darkness.
- Delay of obesity occurred at both short regimens, so aftereffect of one of them does not explain whole result.
- In this model time to development of obesity depended on how external light schedule matched internal rhythm of mice.
We need to double asterisk key numbers, percentages, drug/company/gene names, dates. So we need to select at most 4-5 facts to bold.
Potential bold items: 4.0 ± 1.1 weeks, 12.1 ± 1.6 weeks, 11.3 ± 1.6 weeks, about eight weeks, September 1, maybe tau? But tau is a term not a number. Could bold the numbers: 4.0 ± 1.1 weeks, 12.1 ± 1.6 weeks, 11.3 ± 1.6 weeks, about eight weeks, September 1. That's 5 items.
We must not bold whole sentences; just wrap the numbers/dates.
We also need to wrap study/journal citation in single underscores: The FASEB Journal, September 1.
Now format: headline line, blank line, then paragraphs.
Headline: something like "Shorter Light-Dark Cycles Delay Obesity Onset in Mice". Under 90 chars.
Now body: we need short paragraphs 2-3 sentences each.
Let's craft maybe 4 paragraphs.
Paragraph 1: Introduce study, mention journal and date, mice, high-fat diet, three light-dark cycles.
Paragraph 2: Provide obesity onset times for each cycle, note similar energy intake.
Paragraph 3: Describe activity differences and link to energy expenditure, mention tau observations.
Paragraph 4: Summarize conclusion about external schedule matching internal rhythm.
Make sure to bold numbers/dates.
Double asterisk formatting: 4.0 ± 1.1 weeks, 12.1 ± 1.6 weeks, 11.3 ± 1.6 weeks, about eight weeks, September 1.…
🔗 Read original →
First line: short headline under 90 chars, no markdown, no '#'.
Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.
Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks. At most 4-5 per post, never a whole sentence.
Wrap study/journal citations and publication references in single underscores (e.g., Nature Aging, July 2026).
We need to preserve facts, numbers, names, citations exactly.
Let's extract facts:
- Study: September 1 in The FASEB Journal (date and journal). So citation: The FASEB Journal, September 1? Actually they said "1 сентября в статье The FASEB Journal". So we need to wrap the citation in single underscores: The FASEB Journal, September 1.
- Researchers compared three light-dark cycles in female mice of one lab strain, fed high-fat diet.
- Each group had eight animals; varied duration of light-dark alternation.
- Mice have internal clocks: in constant darkness activity repeats with its own period called tau.
- Early in experiment tau of mice on high-fat diet was longer than that of mice on low-fat diet, before obesity appeared.
- This observation prompted question: would obesity timing be changed by external light schedule aligned with internal rhythm.
- For testing on high-fat diet they used: usual 24-hour day, cycle about 23.64 hours — close to tau — and cycle 23.28 hours.
- Light and dark parts of each cycle occupied equal time.
- At 24-hour regimen obesity recorded at 4.0 ± 1.1 weeks.
- At two shorter cycles — at 12.1 ± 1.6 and 11.3 ± 1.6 weeks.
- Obesity occurred in all three groups, and shorter days shifted its onset by about eight weeks.
- Accumulated energy intake between light regimens was similar.
- Mice with short cycles maintained higher locomotor activity from second week; body mass differences appeared later.
- Authors link this sequence to higher energy expenditure: timing difference of obesity arose on same high-fat diet and without difference in accumulated energy intake.
- After the shortest regimen, 23.28 hours, internal period in darkness also became shorter.
- After cycle close to tau, it remained at level of the cycle period and constant darkness.
- Delay of obesity occurred at both short regimens, so aftereffect of one of them does not explain whole result.
- In this model time to development of obesity depended on how external light schedule matched internal rhythm of mice.
We need to double asterisk key numbers, percentages, drug/company/gene names, dates. So we need to select at most 4-5 facts to bold.
Potential bold items: 4.0 ± 1.1 weeks, 12.1 ± 1.6 weeks, 11.3 ± 1.6 weeks, about eight weeks, September 1, maybe tau? But tau is a term not a number. Could bold the numbers: 4.0 ± 1.1 weeks, 12.1 ± 1.6 weeks, 11.3 ± 1.6 weeks, about eight weeks, September 1. That's 5 items.
We must not bold whole sentences; just wrap the numbers/dates.
We also need to wrap study/journal citation in single underscores: The FASEB Journal, September 1.
Now format: headline line, blank line, then paragraphs.
Headline: something like "Shorter Light-Dark Cycles Delay Obesity Onset in Mice". Under 90 chars.
Now body: we need short paragraphs 2-3 sentences each.
Let's craft maybe 4 paragraphs.
Paragraph 1: Introduce study, mention journal and date, mice, high-fat diet, three light-dark cycles.
Paragraph 2: Provide obesity onset times for each cycle, note similar energy intake.
Paragraph 3: Describe activity differences and link to energy expenditure, mention tau observations.
Paragraph 4: Summarize conclusion about external schedule matching internal rhythm.
Make sure to bold numbers/dates.
Double asterisk formatting: 4.0 ± 1.1 weeks, 12.1 ± 1.6 weeks, 11.3 ± 1.6 weeks, about eight weeks, September 1.…
🔗 Read original →
PubMed Central (PMC)
Photic Cycle Shorter Than or Equal to Endogenous tau Postpones Diet‐Induced Obesity in Mice and Shows a Robust Aftereffect
High‐fat diet (HFD)‐induced obesity (DIO) is preceded by disruptions in endogenous circadian rhythmicity, including lengthening of its period (tau). We previously demonstrated that housing mice under a light–dark cycle (T‐cycle) oscillating at their ...
European Genomic Data Improves Polygenic Predictions for Japanese When Local Sample Is Small
On September 3, Google Research researchers varied the sizes of European and Japanese training sets and tested predictions for eight biological traits on a held‑out portion of Biobank Japan. They examined how the number of local genomes and the trait itself affected the benefit of an external database.
Polygenic scores aggregate the effects of many DNA variants to forecast a trait or disease risk. Such models are usually trained on large cohorts enriched with European ancestry, and accuracy can drop when applied to other groups because variant frequencies and their relationships to traits differ across populations.
In the experiment the European subset of UK Biobank — a British biobank with genetic and medical data — was compared with Biobank Japan, which contains nearly 200 000 participants. For eight traits, including body‑mass index, blood pressure, blood measures, HDL‑ and LDL‑cholesterol, and glucose, the sizes of the two training sets were altered. All models were then evaluated on a separate slice of Biobank Japan not used in training.
Initially, variants for prediction were sought in the European UK Biobank. With about 5 000 Japanese samples, adding the European cohort improved accuracy by giving the model more observations to capture weak effects. At 15 000 Japanese samples or more, a model trained solely on Japanese data outperformed the combined‑data model. For HDL cholesterol, adding over 5 000 European samples beyond that point reduced accuracy, and the cutoff varied by trait.
For body‑mass index, genetic links were more similar between the populations, so European data helped up to 25–40 000 or more Japanese samples. For HDL, LDL‑cholesterol, and glucose, peak accuracy required fewer European samples even with smaller Japanese cohorts.
The researchers also tested other strategies: meta‑analysis combined results from the two populations during variant discovery, which especially aided small
🔗 Read original →
On September 3, Google Research researchers varied the sizes of European and Japanese training sets and tested predictions for eight biological traits on a held‑out portion of Biobank Japan. They examined how the number of local genomes and the trait itself affected the benefit of an external database.
Polygenic scores aggregate the effects of many DNA variants to forecast a trait or disease risk. Such models are usually trained on large cohorts enriched with European ancestry, and accuracy can drop when applied to other groups because variant frequencies and their relationships to traits differ across populations.
In the experiment the European subset of UK Biobank — a British biobank with genetic and medical data — was compared with Biobank Japan, which contains nearly 200 000 participants. For eight traits, including body‑mass index, blood pressure, blood measures, HDL‑ and LDL‑cholesterol, and glucose, the sizes of the two training sets were altered. All models were then evaluated on a separate slice of Biobank Japan not used in training.
Initially, variants for prediction were sought in the European UK Biobank. With about 5 000 Japanese samples, adding the European cohort improved accuracy by giving the model more observations to capture weak effects. At 15 000 Japanese samples or more, a model trained solely on Japanese data outperformed the combined‑data model. For HDL cholesterol, adding over 5 000 European samples beyond that point reduced accuracy, and the cutoff varied by trait.
For body‑mass index, genetic links were more similar between the populations, so European data helped up to 25–40 000 or more Japanese samples. For HDL, LDL‑cholesterol, and glucose, peak accuracy required fewer European samples even with smaller Japanese cohorts.
The researchers also tested other strategies: meta‑analysis combined results from the two populations during variant discovery, which especially aided small
🔗 Read original →
PubMed Central (PMC)
Current clinical use of polygenic scores will risk exacerbating health disparities
Polygenic risk scores (PRS) are poised to improve biomedical outcomes via precision medicine. However, the major ethical and scientific challenge surrounding clinical implementation is that they are many-fold more accurate in European ancestry ...
Experts Call for Pre‑Trial Planning of Neuro‑Implant Responsibility in Australia
A multidisciplinary panel of 24 experts has produced 11 recommendations for Australian clinical trials of implantable neurodevices. The recommendations were published in a review by The Conversation, September 3. In a final vote, 14 panel members participated and every recommendation received at least 80% support.
Before surgery, participants should be told whether they can keep using the device after the trial, what technical and clinical services will remain available, and under what conditions doctors might advise removal. The authors urge that responsibility for post‑trial device maintenance be identified during study design and ethical review. The treating physician should be engaged from the start, receiving results and future plans while the team handles programming updates and spare parts.
An international survey conducted in 2024 gathered responses from 66 researchers about 65 unique neuro‑implant trials
🔗 Read original →
A multidisciplinary panel of 24 experts has produced 11 recommendations for Australian clinical trials of implantable neurodevices. The recommendations were published in a review by The Conversation, September 3. In a final vote, 14 panel members participated and every recommendation received at least 80% support.
Before surgery, participants should be told whether they can keep using the device after the trial, what technical and clinical services will remain available, and under what conditions doctors might advise removal. The authors urge that responsibility for post‑trial device maintenance be identified during study design and ethical review. The treating physician should be engaged from the start, receiving results and future plans while the team handles programming updates and spare parts.
An international survey conducted in 2024 gathered responses from 66 researchers about 65 unique neuro‑implant trials
🔗 Read original →
PubMed Central (PMC)
Post-trial access to implantable neural devices: an exploratory international survey
Clinical trials of innovative neural implants are rapidly increasing and diversifying, but little is known about participants’ post-trial access to the device and ongoing clinical care. This exploratory study examines common practices in the ...
Cure Platform Updates Longevity Biotech Deal Tracker for 2026
On September 1, the Cure platform updated its tracker of eight longevity‑biotech deals for 2026, posting eight funding events on its page. The tracker places investment rounds and non‑dilutive funding — such as grants where the company does not give equity — side by side, and notes each program’s nearest planned step. Cure counts both investment rounds and non‑dilutive funding as raised capital, while potential partnership and royalty payouts are listed separately.
Among the entries are $3 million for Reservoir Neuroscience and $435 million for NewLimit, each paired with the program’s next step. In June NewLimit closed its Series C round of $435 million; the company said it is preparing its first drug for human testing in 2027. Its program transiently alters gene activity in old liver cells while preserving their specialization, and the round funds preparation for this first clinical step.
Life Biosciences raised $80 million in a Series D round. In June the first participant received ER‑100, a gene therapy for optic‑nerve diseases that uses partial epigenetic reprogramming to shift gene activity toward a younger state without changing cell type. According to the company’s release, the funds support completion of the first ER‑100 phase, further work on the platform, and operations through the second half of 2027.
The tracker also includes non‑dilutive funding: Nula Therapeutics announced up to $20 million in support, with plans to test NLT‑101 in humans in Q4 2026 and a separate program assessing functional resilience. Cambrian Bio reported up to $30.8 million for developing a drug that selectively inhibits mTORC1, a protein complex that senses nutrients in cells. Cure’s map keeps the amount, funding type, and each program’s nearest announced action together, grouping the eight deals
🔗 Read original →
On September 1, the Cure platform updated its tracker of eight longevity‑biotech deals for 2026, posting eight funding events on its page. The tracker places investment rounds and non‑dilutive funding — such as grants where the company does not give equity — side by side, and notes each program’s nearest planned step. Cure counts both investment rounds and non‑dilutive funding as raised capital, while potential partnership and royalty payouts are listed separately.
Among the entries are $3 million for Reservoir Neuroscience and $435 million for NewLimit, each paired with the program’s next step. In June NewLimit closed its Series C round of $435 million; the company said it is preparing its first drug for human testing in 2027. Its program transiently alters gene activity in old liver cells while preserving their specialization, and the round funds preparation for this first clinical step.
Life Biosciences raised $80 million in a Series D round. In June the first participant received ER‑100, a gene therapy for optic‑nerve diseases that uses partial epigenetic reprogramming to shift gene activity toward a younger state without changing cell type. According to the company’s release, the funds support completion of the first ER‑100 phase, further work on the platform, and operations through the second half of 2027.
The tracker also includes non‑dilutive funding: Nula Therapeutics announced up to $20 million in support, with plans to test NLT‑101 in humans in Q4 2026 and a separate program assessing functional resilience. Cambrian Bio reported up to $30.8 million for developing a drug that selectively inhibits mTORC1, a protein complex that senses nutrients in cells. Cure’s map keeps the amount, funding type, and each program’s nearest announced action together, grouping the eight deals
🔗 Read original →
Wewillcure
Cure’s Longevity Biotech Funding Tracker 2026
The latest longevity biotech funding news, updated monthly—tracking major rounds and non-dilutive awards of $10M or more in 2026.
We need to translate Russian news post into natural English, format per rules.
First line: short headline under 90 chars, no markdown. Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.
We need to wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks. At most 4-5 per post, never a whole sentence.
Wrap study/journal citations and publication references in single underscores (e.g., Nature Aging, July 2026).
We have content: SciLaws-Bench checking AI ability to differentiate scientific law from curve fitting. Authors published preprint SciLaws-Bench on September 1: 118 tasks gathered from 381 scientific articles. In each they test model in two modes — with real data and in artificial "parallel" world. One measurement table may fit several formulas. One better predicts hidden part of data, another also satisfies task requirements: preserves required sign of quantity, known limits, mandatory dependencies. SciLaws-Bench separates these two abilities. In SciLaws-Real system receives real observations. Its formula is tested on held-out data and by rules from original article: does it preserve required sign, limits, mandatory dependencies. These points evaluated by language model judge; authors compared its decisions with evaluations of five domain experts. In SciLaws-Parallel authors construct artificial world with new version of published formula. System starts without observations, itself selects measurement points, receives noisy answers, and recovers structure of hidden law. Thus test checks selection of measurements that distinguish competing formulas. In authors' test of nine models, more accurate fitting coincided with higher scientific validity only in 54.9% of 3,616 pairwise comparisons. In nuclear physics task, best formula by accuracy added a resonance peak not present in original phenomenon. "Models are able to generate laws better than to reliably select them," write authors.
We need to preserve facts, numbers, names, citations exactly. There's no explicit journal citation; it's a preprint. Could treat "preprint SciLaws-Bench" as maybe a citation? Not sure. The rule: wrap study/journal citations and publication references in single underscores. We have no explicit journal; maybe we can treat "preprint SciLaws-Bench" as a reference? Could wrap "SciLaws-Bench preprint"? But that might be considered a citation. However we need to be careful: we should not add invented detail. The original says "1 сентября авторы опубликовали препринт SciLaws-Bench". So we can mention "On September 1, the authors released the SciLaws-Bench preprint." The phrase "SciLaws-Bench" is a name; we might not need to wrap it. The rule says wrap study/journal citations and publication references in single underscores. The preprint is a publication reference; we could wrap "SciLaws-Bench preprint". That seems appropriate.
Important facts to bold: numbers: September 1, 118 tasks, 381 scientific articles, two modes, 54.9%, 3,616 pairwise comparisons, nine models, nuclear physics task. Also maybe "54.9%" is a percentage. We need at most 4-5 double asterisks. Choose key numbers: September 1, 118 tasks, 381 articles, 54.9%, 3,616 comparisons. That's 5. Could also include "nine models" but that would exceed. We'll stick with those.
We must not bold whole sentences. So we need to embed bold within sentences.
Now format: Headline line (under 90 chars). Something like: "SciLaws-Bench tests AI’s ability to tell real scientific laws from curve fits". Count characters: Let's count: "SciLaws-Bench tests AI’s ability to tell real scientific laws from curve fits". That's about 66 characters. Good.
Then blank line.
Then body paragraphs, each 2-3 sentences.
We need to include bolded numbers and underscores for citation.…
🔗 Read original →
First line: short headline under 90 chars, no markdown. Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.
We need to wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks. At most 4-5 per post, never a whole sentence.
Wrap study/journal citations and publication references in single underscores (e.g., Nature Aging, July 2026).
We have content: SciLaws-Bench checking AI ability to differentiate scientific law from curve fitting. Authors published preprint SciLaws-Bench on September 1: 118 tasks gathered from 381 scientific articles. In each they test model in two modes — with real data and in artificial "parallel" world. One measurement table may fit several formulas. One better predicts hidden part of data, another also satisfies task requirements: preserves required sign of quantity, known limits, mandatory dependencies. SciLaws-Bench separates these two abilities. In SciLaws-Real system receives real observations. Its formula is tested on held-out data and by rules from original article: does it preserve required sign, limits, mandatory dependencies. These points evaluated by language model judge; authors compared its decisions with evaluations of five domain experts. In SciLaws-Parallel authors construct artificial world with new version of published formula. System starts without observations, itself selects measurement points, receives noisy answers, and recovers structure of hidden law. Thus test checks selection of measurements that distinguish competing formulas. In authors' test of nine models, more accurate fitting coincided with higher scientific validity only in 54.9% of 3,616 pairwise comparisons. In nuclear physics task, best formula by accuracy added a resonance peak not present in original phenomenon. "Models are able to generate laws better than to reliably select them," write authors.
We need to preserve facts, numbers, names, citations exactly. There's no explicit journal citation; it's a preprint. Could treat "preprint SciLaws-Bench" as maybe a citation? Not sure. The rule: wrap study/journal citations and publication references in single underscores. We have no explicit journal; maybe we can treat "preprint SciLaws-Bench" as a reference? Could wrap "SciLaws-Bench preprint"? But that might be considered a citation. However we need to be careful: we should not add invented detail. The original says "1 сентября авторы опубликовали препринт SciLaws-Bench". So we can mention "On September 1, the authors released the SciLaws-Bench preprint." The phrase "SciLaws-Bench" is a name; we might not need to wrap it. The rule says wrap study/journal citations and publication references in single underscores. The preprint is a publication reference; we could wrap "SciLaws-Bench preprint". That seems appropriate.
Important facts to bold: numbers: September 1, 118 tasks, 381 scientific articles, two modes, 54.9%, 3,616 pairwise comparisons, nine models, nuclear physics task. Also maybe "54.9%" is a percentage. We need at most 4-5 double asterisks. Choose key numbers: September 1, 118 tasks, 381 articles, 54.9%, 3,616 comparisons. That's 5. Could also include "nine models" but that would exceed. We'll stick with those.
We must not bold whole sentences. So we need to embed bold within sentences.
Now format: Headline line (under 90 chars). Something like: "SciLaws-Bench tests AI’s ability to tell real scientific laws from curve fits". Count characters: Let's count: "SciLaws-Bench tests AI’s ability to tell real scientific laws from curve fits". That's about 66 characters. Good.
Then blank line.
Then body paragraphs, each 2-3 sentences.
We need to include bolded numbers and underscores for citation.…
🔗 Read original →
We need to translate Russian news to English, format per rules.
First line: short headline under 90 chars, no markdown, no '#'.
Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.
Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks. At most 4-5 per post, never a whole sentence.
Wrap study/journal citations and publication references in single underscores.
We need to preserve facts, numbers, names, citations exactly.
Let's extract facts:
- NovaGo Therapeutics (Swiss biotech) announced closing of financing round on 24 million Swiss francs, or $30 million. Date: 2 September (presumably 2025? Not specified but article says "2 сентября швейцарская биотехнологическая компания NovaGo Therapeutics объявила о закрытии раунда финансирования". Likely 2025? Not given; we keep as "2 September". No year given; we can keep just date.
- Funds intended for study of antibody NG004 in 120 people with acute spinal cord injury in Germany, Spain, Switzerland.
- Participants to be randomly assigned to groups, allocation concealed from patients and investigators (double-blind).
- Acute spinal cord injury damages nerve pathways that brain uses to send commands to body; in CNS such fibers poorly regrow, making recovery of movement and sensation difficult.
- NovaGo links NG004 to one of molecular obstacles to such recovery.
- Nogo-A — protein described as brake on nerve fiber regrowth and remodeling of nerve connections after injury.
- NG004 — antibody designed to block Nogo-A.
- Developers expect blockade will allow surviving pathways to form new connections.
- In international clinical trial registry ClinicalTrials.gov, entry last updated February 2025, described first phase of NG004: first-in-human administration to people with acute incomplete cervical spinal cord injury.
- Protocol of this phase envisages dose escalation and assessment of safety, tolerability, pharmacokinetics — how antibody behaves in body.
- With these funds NovaGo plans to move from early testing of dosing regimen to 120-patient NG004 study.
We need to wrap key numbers, percentages, drug/company/gene names, dates in double asterisks. At most 4-5 per post.
Select important facts: $30 million, 24 million Swiss francs, 2 September, 120 patients, Nogo-A, NG004, ClinicalTrials.gov, February 2025, maybe acute spinal cord injury.
But we must limit to 4-5 double asterisk wrappings. Let's choose: $30 million, 24 million Swiss francs, 2 September, 120 patients, Nogo-A, NG004, ClinicalTrials.gov, February 2025 – that's 8, too many. Need max 4-5.
We can combine some: maybe only wrap numbers and key names: $30 million, 120 patients, Nogo-A, NG004, ClinicalTrials.gov (that's 5). Dates maybe not wrapped? But date is important fact; we could wrap 2 September instead of one of others. Need to be selective.
Let's decide: wrap $30 million, 120 patients, Nogo-A, NG004, ClinicalTrials.gov. That's 5. Dates not wrapped; but we could incorporate date in sentence without wrapping.
Alternatively, we could wrap 2 September instead of one. But we need to keep key numbers: financing amount, patient count, protein name, antibody name, registry. That seems good.…
🔗 Read original →
First line: short headline under 90 chars, no markdown, no '#'.
Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.
Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks. At most 4-5 per post, never a whole sentence.
Wrap study/journal citations and publication references in single underscores.
We need to preserve facts, numbers, names, citations exactly.
Let's extract facts:
- NovaGo Therapeutics (Swiss biotech) announced closing of financing round on 24 million Swiss francs, or $30 million. Date: 2 September (presumably 2025? Not specified but article says "2 сентября швейцарская биотехнологическая компания NovaGo Therapeutics объявила о закрытии раунда финансирования". Likely 2025? Not given; we keep as "2 September". No year given; we can keep just date.
- Funds intended for study of antibody NG004 in 120 people with acute spinal cord injury in Germany, Spain, Switzerland.
- Participants to be randomly assigned to groups, allocation concealed from patients and investigators (double-blind).
- Acute spinal cord injury damages nerve pathways that brain uses to send commands to body; in CNS such fibers poorly regrow, making recovery of movement and sensation difficult.
- NovaGo links NG004 to one of molecular obstacles to such recovery.
- Nogo-A — protein described as brake on nerve fiber regrowth and remodeling of nerve connections after injury.
- NG004 — antibody designed to block Nogo-A.
- Developers expect blockade will allow surviving pathways to form new connections.
- In international clinical trial registry ClinicalTrials.gov, entry last updated February 2025, described first phase of NG004: first-in-human administration to people with acute incomplete cervical spinal cord injury.
- Protocol of this phase envisages dose escalation and assessment of safety, tolerability, pharmacokinetics — how antibody behaves in body.
- With these funds NovaGo plans to move from early testing of dosing regimen to 120-patient NG004 study.
We need to wrap key numbers, percentages, drug/company/gene names, dates in double asterisks. At most 4-5 per post.
Select important facts: $30 million, 24 million Swiss francs, 2 September, 120 patients, Nogo-A, NG004, ClinicalTrials.gov, February 2025, maybe acute spinal cord injury.
But we must limit to 4-5 double asterisk wrappings. Let's choose: $30 million, 24 million Swiss francs, 2 September, 120 patients, Nogo-A, NG004, ClinicalTrials.gov, February 2025 – that's 8, too many. Need max 4-5.
We can combine some: maybe only wrap numbers and key names: $30 million, 120 patients, Nogo-A, NG004, ClinicalTrials.gov (that's 5). Dates maybe not wrapped? But date is important fact; we could wrap 2 September instead of one of others. Need to be selective.
Let's decide: wrap $30 million, 120 patients, Nogo-A, NG004, ClinicalTrials.gov. That's 5. Dates not wrapped; but we could incorporate date in sentence without wrapping.
Alternatively, we could wrap 2 September instead of one. But we need to keep key numbers: financing amount, patient count, protein name, antibody name, registry. That seems good.…
🔗 Read original →
PR Newswire
NovaGo Therapeutics Closes USD 30 Million Series B Financing to Advance Proof-of-Concept Study in Acute Spinal Cord Injury
/PRNewswire/ -- NovaGo Therapeutics AG, a clinical-stage biotechnology company developing anti-Nogo-A biologics for diseases of the central nervous system,...
OpenAI warns chain‑of‑thought tracking of Astra is fragile amid depth limits
On 2 September, chief scientist Jakub Pachocki of OpenAI said that the sequential computation depth of the advanced model Astra stays within twice the depth of GPT‑4. He noted that this limit reflects the current frontier of the model’s reasoning steps.
AI safety researcher Ryan Greenblatt asked whether developers could quickly raise the number of internal steps the model takes. OpenAI tracks the model’s chain‑of‑thought text—the written trace of its reasoning—to see how training‑learned behavior rules appear on new tasks.
Pachocki called this monitoring method fragile and said it is moving in an unfavorable direction. In earlier descriptions he portrayed Astra as a system for long‑term research that can modify code, run experiments, and report results.
Greenblatt considers the current disclosure of depth useful and defines sequential depth as the number of computation stages the model passes through one after another before answering. He wonders whether Astra has an adjustable parameter that could increase repetitions, how easy it would be to scale, whether it would boost capabilities, and if a small amount of fine‑tuning could sharply raise the repetition
🔗 Read original →
On 2 September, chief scientist Jakub Pachocki of OpenAI said that the sequential computation depth of the advanced model Astra stays within twice the depth of GPT‑4. He noted that this limit reflects the current frontier of the model’s reasoning steps.
AI safety researcher Ryan Greenblatt asked whether developers could quickly raise the number of internal steps the model takes. OpenAI tracks the model’s chain‑of‑thought text—the written trace of its reasoning—to see how training‑learned behavior rules appear on new tasks.
Pachocki called this monitoring method fragile and said it is moving in an unfavorable direction. In earlier descriptions he portrayed Astra as a system for long‑term research that can modify code, run experiments, and report results.
Greenblatt considers the current disclosure of depth useful and defines sequential depth as the number of computation stages the model passes through one after another before answering. He wonders whether Astra has an adjustable parameter that could increase repetitions, how easy it would be to scale, whether it would boost capabilities, and if a small amount of fine‑tuning could sharply raise the repetition
🔗 Read original →
X (formerly Twitter)
Jakub Pachocki (@merettm) on X
I want to prevent a race into unmonitorability kicked off by confused reporting. The depth of the computation graph for our present frontier models, including Astra, is within a factor of two of G…
We need to translate Russian news post into English, format per rules.
First line: short headline under 90 chars, no markdown, no '#'.
Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.
Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks . At most 4-5 per post, never a whole sentence.
Wrap study/journal citations and publication references in single underscores _ (e.g., Nature Aging, July 2026).
We must preserve all facts, numbers, names, citations exactly. No added commentary.
We need to identify important facts: numbers: 2 September (date), Nature journal, 105,502 gene activity profiles, 10 donors with MS and 10 controls, maybe percentages? Not given. Gene names: CFH, Cfh, C3, SCR20, mHDM-FH. Also maybe mention "oxidative damage", "lipid peroxidation". But we need to limit double asterisks to at most 4-5. Choose key numbers/dates: 2 September, Nature, 105,502, CFH, Cfh, C3, SCR20, mHDM-FH maybe too many. Need max 4-5. Let's pick: 2 September, Nature, 105,502, CFH, Cfh? That's 5. Could also include C3 but that would be 6. Maybe we can combine some: we need to keep key facts but limited. Perhaps we can include 2 September, Nature, 105,502, CFH/Cfh, C3? That's still 5 if we treat CFH/Cfh as one? But they are separate names. Could we wrap gene names in double asterisks only when they appear first time? Might exceed limit. Let's think: The instruction: "Wrap the few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks. Be selective: at most 4-5 per post, never a whole sentence."
Thus we need to choose up to 4-5 items to bold. Likely we can bold: 2 September, Nature, 105,502, CFH, C3. That's 5. Or we could bold CFH, Cfh, C3, SCR20, mHDM-FH (5 gene names) but then we lose date/journal. Probably date and journal are important. Let's choose: 2 September, Nature, 105,502, CFH, C3. That covers date, journal, sample size, key gene, complement gene.
We'll need to ensure we don't bold whole sentences. Just wrap those specific tokens.
Now citations: we have only one citation: article published in Nature on 2 September. So we wrap that in single underscores: Nature maybe with date? The rule: wrap study/journal citations and publication references in single underscores (e.g., Nature Aging, July 2026). So we should put Nature, 2 September maybe. The original Russian: "В опытах на клетках и мышах фактор H защищал нейроны от воспалительного повреждения 2 сентября в журнале Nature вышла статья о факторе H". So citation: Nature, 2 September. We'll wrap that as Nature, 2 September.
Now body: need short paragraphs 2-3 sentences each.
Let's outline translation:
Paragraph 1: In experiments on cells and mice, factor H protected neurons from inflammatory damage. A paper published in Nature, 2 September describes factor H — a protein that normally acts extracellularly as a regulator of the innate immune system. In human post‑mortem tissue, the CFH gene encoding this protein was more active in more resilient neurons.
Paragraph 2: The authors tested how factor H works inside neurons using cell cultures and mice. Protection was provided by its SCR20 domain, which reduced oxidative damage to membranes. Retinal ganglion cells — neurons whose axons form the optic nerve — were isolated from the macula of ten donors with multiple sclerosis and ten control donors; after quality control, 105,502 gene‑expression profiles were analysed.…
🔗 Read original →
First line: short headline under 90 chars, no markdown, no '#'.
Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.
Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks . At most 4-5 per post, never a whole sentence.
Wrap study/journal citations and publication references in single underscores _ (e.g., Nature Aging, July 2026).
We must preserve all facts, numbers, names, citations exactly. No added commentary.
We need to identify important facts: numbers: 2 September (date), Nature journal, 105,502 gene activity profiles, 10 donors with MS and 10 controls, maybe percentages? Not given. Gene names: CFH, Cfh, C3, SCR20, mHDM-FH. Also maybe mention "oxidative damage", "lipid peroxidation". But we need to limit double asterisks to at most 4-5. Choose key numbers/dates: 2 September, Nature, 105,502, CFH, Cfh, C3, SCR20, mHDM-FH maybe too many. Need max 4-5. Let's pick: 2 September, Nature, 105,502, CFH, Cfh? That's 5. Could also include C3 but that would be 6. Maybe we can combine some: we need to keep key facts but limited. Perhaps we can include 2 September, Nature, 105,502, CFH/Cfh, C3? That's still 5 if we treat CFH/Cfh as one? But they are separate names. Could we wrap gene names in double asterisks only when they appear first time? Might exceed limit. Let's think: The instruction: "Wrap the few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks. Be selective: at most 4-5 per post, never a whole sentence."
Thus we need to choose up to 4-5 items to bold. Likely we can bold: 2 September, Nature, 105,502, CFH, C3. That's 5. Or we could bold CFH, Cfh, C3, SCR20, mHDM-FH (5 gene names) but then we lose date/journal. Probably date and journal are important. Let's choose: 2 September, Nature, 105,502, CFH, C3. That covers date, journal, sample size, key gene, complement gene.
We'll need to ensure we don't bold whole sentences. Just wrap those specific tokens.
Now citations: we have only one citation: article published in Nature on 2 September. So we wrap that in single underscores: Nature maybe with date? The rule: wrap study/journal citations and publication references in single underscores (e.g., Nature Aging, July 2026). So we should put Nature, 2 September maybe. The original Russian: "В опытах на клетках и мышах фактор H защищал нейроны от воспалительного повреждения 2 сентября в журнале Nature вышла статья о факторе H". So citation: Nature, 2 September. We'll wrap that as Nature, 2 September.
Now body: need short paragraphs 2-3 sentences each.
Let's outline translation:
Paragraph 1: In experiments on cells and mice, factor H protected neurons from inflammatory damage. A paper published in Nature, 2 September describes factor H — a protein that normally acts extracellularly as a regulator of the innate immune system. In human post‑mortem tissue, the CFH gene encoding this protein was more active in more resilient neurons.
Paragraph 2: The authors tested how factor H works inside neurons using cell cultures and mice. Protection was provided by its SCR20 domain, which reduced oxidative damage to membranes. Retinal ganglion cells — neurons whose axons form the optic nerve — were isolated from the macula of ten donors with multiple sclerosis and ten control donors; after quality control, 105,502 gene‑expression profiles were analysed.…
🔗 Read original →
Nature
Intracellular complement factor H protects neurons during CNS inflammation
Nature - CFH is a key mediator of neuronal resilience across the central nervous system in mice and humans.
We need to translate Russian news post into natural English, format per rules.
First line: headline under 90 chars, no markdown. Then blank line, then body paragraphs (2-3 sentences each) separated by blank lines.
We need to wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks. At most 4-5 per post, never a whole sentence.
Wrap study/journal citations and publication references in single underscores (e.g. Nature Aging, July 2026). There's no explicit journal citation in the text, but we have "III фазы INTerpath-001" and "III фазы" etc. Possibly we can treat the study as a citation? The instruction: wrap study/journal citations and publication references in single underscores. We have no explicit journal name, but we could treat "III фазы INTerpath-001" as a study reference? Might be considered a study. Could put INTerpath-001? But they said e.g. Nature Aging, July 2026. So we need to wrap study/journal citations and publication references. The text mentions "III фазы INTerpath-001". That's a trial name. Could be considered a study reference. We'll wrap it in single underscores: INTerpath-001. Also maybe "III фазы" is not needed. We'll just underscore the trial name.
We need to wrap key numbers, percentages, drug/company/gene names, dates in double asterisks. At most 4-5 per post.
Important facts: dates: 2 сентября (September 2), 19 августа (August 19). Drug/company/gene names: MAGE-A3, pембролизумаб (pembrolizumab), intismeran autogene (personalized mRNA therapy), Merck, Moderna. Also maybe "IIB–IV" stages. Also "III фазы". Also maybe "персональная мРНК-терапии intismeran autogene". Also "пембролизумаб". Also "MAGE-A3". Also "Merck and Moderna". Also "INTerpath-001". Also "IIB–IV". Also "September 2" and "August 19". Need to pick at most 4-5 double asterisk items. Let's select: September 2, August 19, MAGE-A3, pembrolizumab, INTerpath-001 (or Merck and Moderna). But we need to keep at most 4-5. Let's do 5: September 2, August 19, MAGE-A3, pembrolizumab, INTerpath-001. Also we could include Merck and Moderna but that would be extra. We'll stick to 5.
Now we need to ensure we don't wrap whole sentences. Just wrap those specific tokens.
Now produce headline: maybe "Essay explains how tumor mutations lead to personalized cancer vaccines". Under 90 chars.
Let's count: "Essay explains how tumor mutations lead to personalized cancer vaccines" length: count characters: E(1) ... Let's approximate: "Essay explains how tumor mutations lead to personalized cancer vaccines" = 58? Actually let's count: Essay(5) + space=1 =>6, explains(7)=>13, space=1=>14, how(3)=>17, space=1=>18, tumor(5)=>23, space=1=>24, mutations(9)=>33, space=1=>34, lead(4)=>38, space=1=>39, to(2)=>41, space=1=>42, personalized(12)=>54, space=1=>55, cancer(6)=>61, space=1=>62, vaccines(7)=>69. So 69 chars, okay.
Now blank line, then body paragraphs.
We need to split into short paragraphs 2-3 sentences each.
Let's craft paragraphs:
Paragraph 1: Introduce essay, author, date, topic.
Sentence: On September 2, machine‑learning engineer Abhishek Mahajan at biotech company published an essay on therapeutic cancer vaccines. He contrasted early attempts to target shared tumor antigens with personalized mRNA drugs made for each patient. Researchers have long sought proteins present in tumors across individuals, such as MAGE‑A3, which was tested in phase III trials for melanoma and non‑small‑cell lung cancer but failed to improve disease‑free survival.
We need to wrap MAGE‑A3 with double asterisks, September 2 with double asterisks. Also maybe "phase III" not needed. Also "September 2" we already have.
Paragraph 2: Explain how T cells recognize peptides via MHC, process to select neoantigens.…
🔗 Read original →
First line: headline under 90 chars, no markdown. Then blank line, then body paragraphs (2-3 sentences each) separated by blank lines.
We need to wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks. At most 4-5 per post, never a whole sentence.
Wrap study/journal citations and publication references in single underscores (e.g. Nature Aging, July 2026). There's no explicit journal citation in the text, but we have "III фазы INTerpath-001" and "III фазы" etc. Possibly we can treat the study as a citation? The instruction: wrap study/journal citations and publication references in single underscores. We have no explicit journal name, but we could treat "III фазы INTerpath-001" as a study reference? Might be considered a study. Could put INTerpath-001? But they said e.g. Nature Aging, July 2026. So we need to wrap study/journal citations and publication references. The text mentions "III фазы INTerpath-001". That's a trial name. Could be considered a study reference. We'll wrap it in single underscores: INTerpath-001. Also maybe "III фазы" is not needed. We'll just underscore the trial name.
We need to wrap key numbers, percentages, drug/company/gene names, dates in double asterisks. At most 4-5 per post.
Important facts: dates: 2 сентября (September 2), 19 августа (August 19). Drug/company/gene names: MAGE-A3, pембролизумаб (pembrolizumab), intismeran autogene (personalized mRNA therapy), Merck, Moderna. Also maybe "IIB–IV" stages. Also "III фазы". Also maybe "персональная мРНК-терапии intismeran autogene". Also "пембролизумаб". Also "MAGE-A3". Also "Merck and Moderna". Also "INTerpath-001". Also "IIB–IV". Also "September 2" and "August 19". Need to pick at most 4-5 double asterisk items. Let's select: September 2, August 19, MAGE-A3, pembrolizumab, INTerpath-001 (or Merck and Moderna). But we need to keep at most 4-5. Let's do 5: September 2, August 19, MAGE-A3, pembrolizumab, INTerpath-001. Also we could include Merck and Moderna but that would be extra. We'll stick to 5.
Now we need to ensure we don't wrap whole sentences. Just wrap those specific tokens.
Now produce headline: maybe "Essay explains how tumor mutations lead to personalized cancer vaccines". Under 90 chars.
Let's count: "Essay explains how tumor mutations lead to personalized cancer vaccines" length: count characters: E(1) ... Let's approximate: "Essay explains how tumor mutations lead to personalized cancer vaccines" = 58? Actually let's count: Essay(5) + space=1 =>6, explains(7)=>13, space=1=>14, how(3)=>17, space=1=>18, tumor(5)=>23, space=1=>24, mutations(9)=>33, space=1=>34, lead(4)=>38, space=1=>39, to(2)=>41, space=1=>42, personalized(12)=>54, space=1=>55, cancer(6)=>61, space=1=>62, vaccines(7)=>69. So 69 chars, okay.
Now blank line, then body paragraphs.
We need to split into short paragraphs 2-3 sentences each.
Let's craft paragraphs:
Paragraph 1: Introduce essay, author, date, topic.
Sentence: On September 2, machine‑learning engineer Abhishek Mahajan at biotech company published an essay on therapeutic cancer vaccines. He contrasted early attempts to target shared tumor antigens with personalized mRNA drugs made for each patient. Researchers have long sought proteins present in tumors across individuals, such as MAGE‑A3, which was tested in phase III trials for melanoma and non‑small‑cell lung cancer but failed to improve disease‑free survival.
We need to wrap MAGE‑A3 with double asterisks, September 2 with double asterisks. Also maybe "phase III" not needed. Also "September 2" we already have.
Paragraph 2: Explain how T cells recognize peptides via MHC, process to select neoantigens.…
🔗 Read original →
Deleting the Ghr gene in mice at 12 months of age using a tamoxifen‑inducible Cre system extended lifespan in females. The animals received tamoxifen for five days, while controls got peanut oil. This lowered circulating IGF-1 while raising growth hormone levels, indicating reduced tissue sensitivity to the signal.
Female mice lived longer: median lifespan rose by +8%, and the maximum recorded lifespan increased by +12%. Male mice did not show a significant shift in survival curves.
In males, after adjusting for lean body mass, performance on the rotarod and grip‑strength test improved, and vertebral trabecular bone structure was better preserved. Liver single‑nucleus profiling showed hepatocyte gene programs shifting toward a more female‑like pattern.
The work was published 5 September and shows that attenuating GH signaling after midlife can still influence aging, with sex‑specific effects on survival and liver metabolism.
🔗 Read original →
Female mice lived longer: median lifespan rose by +8%, and the maximum recorded lifespan increased by +12%. Male mice did not show a significant shift in survival curves.
In males, after adjusting for lean body mass, performance on the rotarod and grip‑strength test improved, and vertebral trabecular bone structure was better preserved. Liver single‑nucleus profiling showed hepatocyte gene programs shifting toward a more female‑like pattern.
The work was published 5 September and shows that attenuating GH signaling after midlife can still influence aging, with sex‑specific effects on survival and liver metabolism.
🔗 Read original →
PubMed Central (PMC)
Midlife Growth Hormone Receptor Ablation Extends Healthy Lifespan and Induces Sex‐Specific Hepatic Transcriptional Changes at Single‐Cell…
Suppression of growth hormone (GH) signaling is known to be effective to extend lifespan in mammals, yet most models rely on congenital disruption of the GH/insulin‐like growth factor‐1 (IGF‐1) axis. Whether modulation of this pathway later in life ...