Mol-JEPA predicts drug‑like molecule properties from structure and biology
Mol‑JEPA, a model for forecasting the properties of medicinal molecules, was released as a preprint on 23 August. The authors compiled data for 4.69 million small molecules, encompassing 14 types of information ranging from chemical structure to cellular measurements and ADMET profiles (absorption, distribution, metabolism, excretion, and toxicity).
During training, Mol‑JEPA randomly masks one data type and learns to reconstruct its compressed numerical description from the remaining modalities. This forces the model to link a molecule’s structure with its biological and pharmacological observations without distorting the underlying structure.
The approach addresses cases where a candidate binds strongly to a target protein but is rapidly cleared or toxic, and where minor structural tweaks can dramatically alter ADMET outcomes. By preserving the exact structure and training on multiple observation types, Mol‑JEPA learns a unified representation that captures diverse molecular behaviors.
Mol‑JEPA encodes each data type—atomic bond schemata, calculated chemical features, biological assay results, cellular profiles, and ADMET data—into a short numerical vector of uniform format. A shared module then receives the available vectors and predicts the representation of the masked type, thereby training the model to connect structure with function.
In property‑prediction tasks on small subsets, Mol‑JEPA achieved lower mean absolute error than competing methods, and its advantage widened as test molecules became more structurally distinct from the training set. On public temporal splits of the TabICLv2 benchmark (a tabular‑data model), Mol‑JEPA was sometimes outperformed in both MAE and head‑to‑head comparisons. Random splits can place molecules sharing a core scaffold in both training and test, leading to overly optimistic estimates; the authors note that evaluating on structurally distant compounds provides a stricter test of transferability.
An ablation study showed that adding more information improves accuracy: a version trained only on molecular graphs and ECFP4 fingerprints was outperformed by the full model using all data types, which reduced MAE by 14 % on a simple task and 13 % on a more complex one.
For drug development, linking molecular structure to accumulated behavior data is essential. Mol‑JEPA consolidates this information into a single numerical descriptor that can be used to predict the properties of new drug candidates.
🔗 Read original →
Mol‑JEPA, a model for forecasting the properties of medicinal molecules, was released as a preprint on 23 August. The authors compiled data for 4.69 million small molecules, encompassing 14 types of information ranging from chemical structure to cellular measurements and ADMET profiles (absorption, distribution, metabolism, excretion, and toxicity).
During training, Mol‑JEPA randomly masks one data type and learns to reconstruct its compressed numerical description from the remaining modalities. This forces the model to link a molecule’s structure with its biological and pharmacological observations without distorting the underlying structure.
The approach addresses cases where a candidate binds strongly to a target protein but is rapidly cleared or toxic, and where minor structural tweaks can dramatically alter ADMET outcomes. By preserving the exact structure and training on multiple observation types, Mol‑JEPA learns a unified representation that captures diverse molecular behaviors.
Mol‑JEPA encodes each data type—atomic bond schemata, calculated chemical features, biological assay results, cellular profiles, and ADMET data—into a short numerical vector of uniform format. A shared module then receives the available vectors and predicts the representation of the masked type, thereby training the model to connect structure with function.
In property‑prediction tasks on small subsets, Mol‑JEPA achieved lower mean absolute error than competing methods, and its advantage widened as test molecules became more structurally distinct from the training set. On public temporal splits of the TabICLv2 benchmark (a tabular‑data model), Mol‑JEPA was sometimes outperformed in both MAE and head‑to‑head comparisons. Random splits can place molecules sharing a core scaffold in both training and test, leading to overly optimistic estimates; the authors note that evaluating on structurally distant compounds provides a stricter test of transferability.
An ablation study showed that adding more information improves accuracy: a version trained only on molecular graphs and ECFP4 fingerprints was outperformed by the full model using all data types, which reduced MAE by 14 % on a simple task and 13 % on a more complex one.
For drug development, linking molecular structure to accumulated behavior data is essential. Mol‑JEPA consolidates this information into a single numerical descriptor that can be used to predict the properties of new drug candidates.
🔗 Read original →
Nature
Impact of molecular multimodality on neural network models for prediction tasks related to drug discovery
Nature Communications - Here, the authors examine whether combining multiple molecular data types improves drug discovery models, finding multimodal approaches boost predictions with effective...
Magnetite nanodiscs trigger astrocyte‑specific calcium response in mouse hippocampus
On 19 August the authors posted a preprint describing an experiment in which mice received magnetite nanodiscs coated with an antibody that binds the astrocyte surface protein GLAST, injected into the dentate gyrus of the hippocampus.
Applying alternating magnetic fields of 25–28–50 mT produced a statistically larger intracellular calcium response in astrocytes than hematite particles of the same shape; the calcium signal was monitored by fiber photometry of a genetically encoded calcium‑sensitive reporter.
Between the second and fifth week after a single injection, magnetite discs elicited a stronger astrocyte response than the hematite control, whereas mice expressing the reporter in neurons showed no difference between the two materials.
The method builds on earlier work from 2022 that used antibody‑functionalized magnetic particles moved by strong field gradients and a 2025 open coil for weak alternating fields developed by Po‑Han Chang’s group; three Science studies have linked astrocyte calcium signals to neural network state switching, allowing causal tests.
🔗 Read original →
On 19 August the authors posted a preprint describing an experiment in which mice received magnetite nanodiscs coated with an antibody that binds the astrocyte surface protein GLAST, injected into the dentate gyrus of the hippocampus.
Applying alternating magnetic fields of 25–28–50 mT produced a statistically larger intracellular calcium response in astrocytes than hematite particles of the same shape; the calcium signal was monitored by fiber photometry of a genetically encoded calcium‑sensitive reporter.
Between the second and fifth week after a single injection, magnetite discs elicited a stronger astrocyte response than the hematite control, whereas mice expressing the reporter in neurons showed no difference between the two materials.
The method builds on earlier work from 2022 that used antibody‑functionalized magnetic particles moved by strong field gradients and a 2025 open coil for weak alternating fields developed by Po‑Han Chang’s group; three Science studies have linked astrocyte calcium signals to neural network state switching, allowing causal tests.
🔗 Read original →
PubMed Central (PMC)
Remote and Selective Control of Astrocytes by Magnetomechanical Stimulation
Astrocytes play crucial and diverse roles in brain health and disease. The ability to selectively control astrocytes provides a valuable tool for understanding their function and has the therapeutic potential to correct dysfunction. Existing ...
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 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 all facts, numbers, names, citations exactly. No added commentary.
We need to identify important facts: DARPA preparing Firefox program for 100-channel neurointerface without surgery, date: September 1 2026 meeting. Also: goal: system reads cortex signals with light through intact skull. Noninvasive coherent optical system already developed records neural signals at cortical depth in humans. Firefox to expand to multi-channel interface. Agency set path to 100 channels, spatial resolution 50 micrometers, temporal 100 microseconds. Program 36 months, three yearly phases; results checked by independent partner. Firefox divided into two related engineering tasks: dynamic scanning to steer light to several brain areas; other task: process large measurement stream: seeking architectures beyond ordinary digital data processing. Same physical difficulty as completed program N3: light, sound, electromagnetic signals scatter and attenuate passing through skin, skull, brain tissue. N3 set benchmark: work with 16 independent channels in volume of neural tissue 16 mm³ and temporal accuracy up to 50 milliseconds. Firefox builds on this problem and sets new measurement parameters.
We need to wrap key numbers/dates etc in . Choose maybe: September 1 2026, 100 channels, 50 micrometers, 100 microseconds, 36 months, three yearly phases, 16 channels, 16 mm³, 50 milliseconds. That's many; we need at most 4-5. So we need to be selective: maybe highlight date, 100 channels, spatial resolution 50 µm, temporal resolution 100 µs, program duration 36 months. That's 5. Or we could include N3 benchmark as well but limit.
We need to wrap study/journal citations in single underscores. There are none explicit citations like journal names. The text mentions "уведомлении о Firefox DARPA пишет" but no journal. So maybe none. We can leave none.
We must not wrap whole sentence. So we need to embed double asterisks around the selected facts within sentences.
We need to ensure not to exceed 4-5 double asterisk occurrences. Each occurrence is a pair .... So we need max 5 pairs.
Let's pick: September 1, 2026, 100 channels, 50 µm spatial resolution, 100 µs temporal resolution, 36‑month program. That's 5.
Make sure to keep numbers exactly as in source: "100 каналов", "50 микрометров", "100 микросекунд", "36 месяцев". Use appropriate formatting.
Now produce headline: maybe "DARPA Launches Firefox Program for 100‑Channel Non‑Invasive Neurointerface". Count chars: Let's count: DARPA (5) + space (1) =6, Launches (8) =>14, space=15, Firefox (7)=>22, space=23, Program (6)=>29, space=30, for (3)=>33, space=34, 100‑Channel (12 inc hyphen?) Let's count: "100‑Channel" characters: 100 (3) + ‑ (maybe hyphen) + Channel (7) = 11? Actually "100‑Channel": 3 + 1 + 7 =11. So total 34+11=45, space=46, Non‑Invasive (13? Non‑ (4) + Invasive (8) + hyphen? Actually "Non‑Invasive": Non‑ (4 inc hyphen?) Let's count: N o n ‑ I n v a s i v e = 4 letters + hyphen + 8 =13. So 46+13=59, space=60, Neurointerface (13?) Neurointerface: N e u r o i n t e r f a c e =12? Actually "Neurointerface": Neuro (5) + interface (8) =13. So 60+13=73. Under 90. Good.
Now blank line, then body paragraphs.
We need short paragraphs 2-3 sentences each.
Let's craft maybe 3 paragraphs.
Paragraph 1: mention meeting date and goal.
Paragraph 2: mention existing system, target specs.
Paragraph 3: program duration, phases, tasks, relation to N3.…
🔗 Read original →
First line: short headline under 90 chars, no markdown. 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 all facts, numbers, names, citations exactly. No added commentary.
We need to identify important facts: DARPA preparing Firefox program for 100-channel neurointerface without surgery, date: September 1 2026 meeting. Also: goal: system reads cortex signals with light through intact skull. Noninvasive coherent optical system already developed records neural signals at cortical depth in humans. Firefox to expand to multi-channel interface. Agency set path to 100 channels, spatial resolution 50 micrometers, temporal 100 microseconds. Program 36 months, three yearly phases; results checked by independent partner. Firefox divided into two related engineering tasks: dynamic scanning to steer light to several brain areas; other task: process large measurement stream: seeking architectures beyond ordinary digital data processing. Same physical difficulty as completed program N3: light, sound, electromagnetic signals scatter and attenuate passing through skin, skull, brain tissue. N3 set benchmark: work with 16 independent channels in volume of neural tissue 16 mm³ and temporal accuracy up to 50 milliseconds. Firefox builds on this problem and sets new measurement parameters.
We need to wrap key numbers/dates etc in . Choose maybe: September 1 2026, 100 channels, 50 micrometers, 100 microseconds, 36 months, three yearly phases, 16 channels, 16 mm³, 50 milliseconds. That's many; we need at most 4-5. So we need to be selective: maybe highlight date, 100 channels, spatial resolution 50 µm, temporal resolution 100 µs, program duration 36 months. That's 5. Or we could include N3 benchmark as well but limit.
We need to wrap study/journal citations in single underscores. There are none explicit citations like journal names. The text mentions "уведомлении о Firefox DARPA пишет" but no journal. So maybe none. We can leave none.
We must not wrap whole sentence. So we need to embed double asterisks around the selected facts within sentences.
We need to ensure not to exceed 4-5 double asterisk occurrences. Each occurrence is a pair .... So we need max 5 pairs.
Let's pick: September 1, 2026, 100 channels, 50 µm spatial resolution, 100 µs temporal resolution, 36‑month program. That's 5.
Make sure to keep numbers exactly as in source: "100 каналов", "50 микрометров", "100 микросекунд", "36 месяцев". Use appropriate formatting.
Now produce headline: maybe "DARPA Launches Firefox Program for 100‑Channel Non‑Invasive Neurointerface". Count chars: Let's count: DARPA (5) + space (1) =6, Launches (8) =>14, space=15, Firefox (7)=>22, space=23, Program (6)=>29, space=30, for (3)=>33, space=34, 100‑Channel (12 inc hyphen?) Let's count: "100‑Channel" characters: 100 (3) + ‑ (maybe hyphen) + Channel (7) = 11? Actually "100‑Channel": 3 + 1 + 7 =11. So total 34+11=45, space=46, Non‑Invasive (13? Non‑ (4) + Invasive (8) + hyphen? Actually "Non‑Invasive": Non‑ (4 inc hyphen?) Let's count: N o n ‑ I n v a s i v e = 4 letters + hyphen + 8 =13. So 46+13=59, space=60, Neurointerface (13?) Neurointerface: N e u r o i n t e r f a c e =12? Actually "Neurointerface": Neuro (5) + interface (8) =13. So 60+13=73. Under 90. Good.
Now blank line, then body paragraphs.
We need short paragraphs 2-3 sentences each.
Let's craft maybe 3 paragraphs.
Paragraph 1: mention meeting date and goal.
Paragraph 2: mention existing system, target specs.
Paragraph 3: program duration, phases, tasks, relation to N3.…
🔗 Read original →
PubMed Central (PMC)
Optical brain imaging in vivo: techniques and applications from animal to man
Optical brain imaging has seen 30 years of intense development, and has grown into a rich and diverse field. In-vivo imaging using light provides unprecedented sensitivity to functional changes through intrinsic contrast, and is rapidly exploiting ...
We need to translate Russian news into English, format with headline line under 90 chars, 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. Wrap study/journal citations and publication references in single underscores.
We need to preserve all facts, numbers, names, citations exactly. No added commentary. No source link.
We need to identify important facts: $40 млн, date 25 августа (August 25), Highland Europe fund, doubling team in Lausanne, opening office with lab in London Q4 2026, Claude campaign 1,320 mini-proteins for 15 targets, Anthropic, Twist Bioscience, first data ~21 days, quote.
We need to wrap key numbers etc in . Choose maybe: $40 million, August 25, Highland Europe, Q4 2026, 1,320 mini-proteins, 15 targets, 21 days. That's 7, but limit 4-5. Need to be selective. Perhaps choose: $40 million, August 25, Q4 2026, 1,320 mini-proteins, 21 days. That's 5. Also need company names? They are important but maybe not wrap. Could wrap Highland Europe? But we already have 5. Could replace one. Let's decide: important facts: funding amount, date, fund name, expansion timeline, Claude campaign size, first data time. That's 6. Need max 5. Could combine fund name with amount? Not. Choose: $40 million, August 25, Highland Europe, Q4 2026, 21 days. Then we omit 1,320 mini-proteins and 15 targets? But they are facts. Could wrap them in double asterisks as well but limit. Maybe we can wrap 1,320 mini-proteins and 15 targets together as one fact? Eg "1,320 mini-proteins across 15 targets". That counts as one wrapped fact. Then we have: $40 million, August 25, Highland Europe, Q4 2026, 1,320 mini-proteins across 15 targets, 21 days => 6. Need 5 max. Could drop fund name? But fund is important. Could drop date? Date is important. Could drop Q4 2026? Maybe less important than others? But expansion timeline is important. Could combine date and fund? Eg "Highland Europe led a $40 million round on August 25". That's one wrapped fact containing fund, amount, date. Then we have: Highland Europe led a $40 million round on August 25, Q4 2026, 1,320 mini-proteins across 15 targets, 21 days. That's 4. Good.
Now we need to wrap study/journal citations in single underscores. There are none explicit besides maybe "Claude campaign" not a journal. No citations. So none.
Now produce headline under 90 chars. Something like: "Adaptive Bio Secures $40 Million to Expand AI‑Driven Protein Lab". Count characters: Let's count: Adaptive(8) + space=1 =>9, Bio=3 =>12, space=13, Secures=7 =>20, space=21, $40=3 =>24, space=25, Million=7 =>32, space=33, to=2 =>35, space=36, Expand=6 =>42, space=43, AI‑Driven=?? "AI‑Driven" includes hyphen and maybe special char. Count characters: A(1) I(2) ‑ (maybe en dash) counts as 1? We'll treat as one. D(4) r(5) i(6) v(7) e(8) n(9) => total 10? Actually "AI‑Driven" length 10 inc hyphen. So add 10 =>53, space=54, Protein=6 =>60, space=61, Lab=3 =>64. So under 90.
Now body paragraphs: need short paragraphs 2-3 sentences each, separated by blank lines.
We'll produce maybe 4 paragraphs.
Paragraph 1: Adaptive Bio announced a $40 million investment round led by Highland Europe on August 25. The funds will double the team in Lausanne and open a London office with a laboratory in Q4 2026.…
🔗 Read original →
We need to preserve all facts, numbers, names, citations exactly. No added commentary. No source link.
We need to identify important facts: $40 млн, date 25 августа (August 25), Highland Europe fund, doubling team in Lausanne, opening office with lab in London Q4 2026, Claude campaign 1,320 mini-proteins for 15 targets, Anthropic, Twist Bioscience, first data ~21 days, quote.
We need to wrap key numbers etc in . Choose maybe: $40 million, August 25, Highland Europe, Q4 2026, 1,320 mini-proteins, 15 targets, 21 days. That's 7, but limit 4-5. Need to be selective. Perhaps choose: $40 million, August 25, Q4 2026, 1,320 mini-proteins, 21 days. That's 5. Also need company names? They are important but maybe not wrap. Could wrap Highland Europe? But we already have 5. Could replace one. Let's decide: important facts: funding amount, date, fund name, expansion timeline, Claude campaign size, first data time. That's 6. Need max 5. Could combine fund name with amount? Not. Choose: $40 million, August 25, Highland Europe, Q4 2026, 21 days. Then we omit 1,320 mini-proteins and 15 targets? But they are facts. Could wrap them in double asterisks as well but limit. Maybe we can wrap 1,320 mini-proteins and 15 targets together as one fact? Eg "1,320 mini-proteins across 15 targets". That counts as one wrapped fact. Then we have: $40 million, August 25, Highland Europe, Q4 2026, 1,320 mini-proteins across 15 targets, 21 days => 6. Need 5 max. Could drop fund name? But fund is important. Could drop date? Date is important. Could drop Q4 2026? Maybe less important than others? But expansion timeline is important. Could combine date and fund? Eg "Highland Europe led a $40 million round on August 25". That's one wrapped fact containing fund, amount, date. Then we have: Highland Europe led a $40 million round on August 25, Q4 2026, 1,320 mini-proteins across 15 targets, 21 days. That's 4. Good.
Now we need to wrap study/journal citations in single underscores. There are none explicit besides maybe "Claude campaign" not a journal. No citations. So none.
Now produce headline under 90 chars. Something like: "Adaptive Bio Secures $40 Million to Expand AI‑Driven Protein Lab". Count characters: Let's count: Adaptive(8) + space=1 =>9, Bio=3 =>12, space=13, Secures=7 =>20, space=21, $40=3 =>24, space=25, Million=7 =>32, space=33, to=2 =>35, space=36, Expand=6 =>42, space=43, AI‑Driven=?? "AI‑Driven" includes hyphen and maybe special char. Count characters: A(1) I(2) ‑ (maybe en dash) counts as 1? We'll treat as one. D(4) r(5) i(6) v(7) e(8) n(9) => total 10? Actually "AI‑Driven" length 10 inc hyphen. So add 10 =>53, space=54, Protein=6 =>60, space=61, Lab=3 =>64. So under 90.
Now body paragraphs: need short paragraphs 2-3 sentences each, separated by blank lines.
We'll produce maybe 4 paragraphs.
Paragraph 1: Adaptive Bio announced a $40 million investment round led by Highland Europe on August 25. The funds will double the team in Lausanne and open a London office with a laboratory in Q4 2026.…
🔗 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 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.
Identify important facts: "26 августа" date, "Рамез Наам", "Дарио Амодеи", "мощный ИИ может дать биологии следующие 100 лет прогресса за 5–10 лет", "четырёх признаках", maybe numbers: 100 years, 5–10 years, four criteria. Also maybe "Go" result immediate, etc. But we need at most 4-5 double asterisks. Choose key numbers: date (26 August), 100 years, 5–10 years, four criteria. That's four.
We need to wrap them in . Ensure not whole sentence.
Citations: There's mention of essay "Machines of Loving Grace" by Amodei. Also maybe mention of review of AI agents for bio labs? Not a journal citation. The essay is a reference; we can wrap in single underscores: Machines of Loving Grace. Also maybe mention of "review of AI agents for bio labs" but not a journal. We'll just underscore the essay title.
We need to preserve all facts exactly; translate.
Let's draft.
Headline: something like "Ramez Naam Links AI Speed in Biology to Organism Response". Under 90 chars.
Now body paragraphs.
Paragraph 1: On August 26, Ramez Naam discussed Dario Amodei's forecast: powerful AI could deliver biology’s next 100 years of progress in just 5–10 years. Naam says the pace depends on how quickly an experiment returns a clear result tied to a real goal.
Paragraph 2: He proposes evaluating a task by four criteria: can success be distinguished from error without mistake; is there enough data; does feedback arrive at computer speed; does the measured outcome match the true objective. In Go the result is immediate; in formal mathematics verification is similar; in programming a test shows code behavior in a given scenario, though human intent may be broader.
Paragraph 3: Fast feedback lets the next attempt be corrected right away. In aging research, cell‑based experiments are only the first step; next we must see if the effect repeats in animals and humans. Naam puts the gap as: “What we get in cells is not the same as what we find in the organism.” A cell living longer in a dish tells us only about cells; to know if an intervention extends human life we must test it on the whole organism.
Paragraph 4: Moving from cells to animals and people changes the measured outcome: a system may improve an early marker while the real question requires a whole‑organism result. In his essay Machines of Loving Grace, Amodei describes AI as a virtual biologist that plans and runs experiments, suggests how to measure outcomes, and directs lab robots or people. He also notes the unavoidable delay of experiments and equipment design.
Paragraph 5: Amodei calculates that smart experiment selection and massive parallelism can hunt for solutions while individual experiments run their course. Naam adds another question: how fast can we go from a cellular result to an answer about humans? If that transition needs long observation of an organism, it sets the learning tempo. In a survey of AI agents for bio labs the cycle looked like: model gathers literature, proposes a hypothesis, sends it to filters and experiment. Naam adds that for this cycle to be fast the experiment must return a signal linked to the desired human outcome.
Now we need to apply formatting: double asterisks around key numbers/dates etc.
Identify key facts to bold:…
🔗 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. No added commentary.
Identify important facts: "26 августа" date, "Рамез Наам", "Дарио Амодеи", "мощный ИИ может дать биологии следующие 100 лет прогресса за 5–10 лет", "четырёх признаках", maybe numbers: 100 years, 5–10 years, four criteria. Also maybe "Go" result immediate, etc. But we need at most 4-5 double asterisks. Choose key numbers: date (26 August), 100 years, 5–10 years, four criteria. That's four.
We need to wrap them in . Ensure not whole sentence.
Citations: There's mention of essay "Machines of Loving Grace" by Amodei. Also maybe mention of review of AI agents for bio labs? Not a journal citation. The essay is a reference; we can wrap in single underscores: Machines of Loving Grace. Also maybe mention of "review of AI agents for bio labs" but not a journal. We'll just underscore the essay title.
We need to preserve all facts exactly; translate.
Let's draft.
Headline: something like "Ramez Naam Links AI Speed in Biology to Organism Response". Under 90 chars.
Now body paragraphs.
Paragraph 1: On August 26, Ramez Naam discussed Dario Amodei's forecast: powerful AI could deliver biology’s next 100 years of progress in just 5–10 years. Naam says the pace depends on how quickly an experiment returns a clear result tied to a real goal.
Paragraph 2: He proposes evaluating a task by four criteria: can success be distinguished from error without mistake; is there enough data; does feedback arrive at computer speed; does the measured outcome match the true objective. In Go the result is immediate; in formal mathematics verification is similar; in programming a test shows code behavior in a given scenario, though human intent may be broader.
Paragraph 3: Fast feedback lets the next attempt be corrected right away. In aging research, cell‑based experiments are only the first step; next we must see if the effect repeats in animals and humans. Naam puts the gap as: “What we get in cells is not the same as what we find in the organism.” A cell living longer in a dish tells us only about cells; to know if an intervention extends human life we must test it on the whole organism.
Paragraph 4: Moving from cells to animals and people changes the measured outcome: a system may improve an early marker while the real question requires a whole‑organism result. In his essay Machines of Loving Grace, Amodei describes AI as a virtual biologist that plans and runs experiments, suggests how to measure outcomes, and directs lab robots or people. He also notes the unavoidable delay of experiments and equipment design.
Paragraph 5: Amodei calculates that smart experiment selection and massive parallelism can hunt for solutions while individual experiments run their course. Naam adds another question: how fast can we go from a cellular result to an answer about humans? If that transition needs long observation of an organism, it sets the learning tempo. In a survey of AI agents for bio labs the cycle looked like: model gathers literature, proposes a hypothesis, sends it to filters and experiment. Naam adds that for this cycle to be fast the experiment must return a signal linked to the desired human outcome.
Now we need to apply formatting: double asterisks around key numbers/dates etc.
Identify key facts to bold:…
🔗 Read original →
PubMed Central (PMC)
Biomarkers of Aging for the Identification and Evaluation of Longevity Interventions
With the rapid expansion of aging biology research, the identification and evaluation of longevity interventions in humans have become key goals of this field. Biomarkers of aging are critically important tools in achieving these objectives over ...
Remote-Controlled Neuroimplant Delivers Light and Drugs Via Internet
Researchers in Chicago demonstrated a neuroimplant that can be operated from anywhere via the internet. They remotely controlled a device implanted in the brain of a laboratory rat located in Daejeon, South Korea. The implant functions in real time, turning on stimulation or drug delivery with a simple web command.
The platform combines optogenetics and microfluidics, allowing neurons to be activated by light or inhibited while drugs are precisely injected. By eliminating bulky tethers and external hardware, it removes the observer effect and reduces stress on the animals.
Inside the chip sits an electrochemical micropump: applying voltage triggers electrolysis, generating gas that pushes the drug from its reservoir; cutting the current lets the gas recombine, resetting the system. Dosing and the frequency of light pulses from a micro-LED are adjusted through a web interface, and the drug cartridge attaches magnetically for easy replacement or refilling without additional surgery.
Stability and durability were validated in a four-week experiment on rats. In addiction studies, scientists remotely delivered cocaine to the Nucleus Accumbens while simultaneously suppressing neural activity with light, successfully curbing addictive behavior. Looking ahead, Korean developers aim to link such IoT implants with AI to close the loop—letting algorithms read brain states and automatically administer the appropriate drug or light pulse for conditions such as depression, neurodegenerative disorders, or epilepsy.
🔗 Read original →
Researchers in Chicago demonstrated a neuroimplant that can be operated from anywhere via the internet. They remotely controlled a device implanted in the brain of a laboratory rat located in Daejeon, South Korea. The implant functions in real time, turning on stimulation or drug delivery with a simple web command.
The platform combines optogenetics and microfluidics, allowing neurons to be activated by light or inhibited while drugs are precisely injected. By eliminating bulky tethers and external hardware, it removes the observer effect and reduces stress on the animals.
Inside the chip sits an electrochemical micropump: applying voltage triggers electrolysis, generating gas that pushes the drug from its reservoir; cutting the current lets the gas recombine, resetting the system. Dosing and the frequency of light pulses from a micro-LED are adjusted through a web interface, and the drug cartridge attaches magnetically for easy replacement or refilling without additional surgery.
Stability and durability were validated in a four-week experiment on rats. In addiction studies, scientists remotely delivered cocaine to the Nucleus Accumbens while simultaneously suppressing neural activity with light, successfully curbing addictive behavior. Looking ahead, Korean developers aim to link such IoT implants with AI to close the loop—letting algorithms read brain states and automatically administer the appropriate drug or light pulse for conditions such as depression, neurodegenerative disorders, or epilepsy.
🔗 Read original →
Science Advances
IoT-enabled wireless neural implant for chronic, programmable neuropharmacology and optogenetics
IoT-enabled wireless implant with refillable drug delivery and optogenetics enables chronic, observer-free neural modulation.
Alex Colvill: Longevity Interest Went Mainstream Before Human Therapies Arrived
On 26 August, in a new episode of the Core Memory podcast, Ashley Vance spoke with Alex Colvill, co‑founder of venture fund age1. Colvill said that interest in longevity had already become widespread even before ready‑to‑use therapies for people are available.
He believes the number of interventions that can be tested in people
🔗 Read original →
On 26 August, in a new episode of the Core Memory podcast, Ashley Vance spoke with Alex Colvill, co‑founder of venture fund age1. Colvill said that interest in longevity had already become widespread even before ready‑to‑use therapies for people are available.
He believes the number of interventions that can be tested in people
🔗 Read original →
Corememory
The Realest Longevity Guru - EP 85 Alex Colville
Over the past five years, Alex Colville has become one of the most influential figures in the longevity and aging fields.
Self‑Training Bioengineered Muscle Implant Improves Aging in Mice
Scientists took mouse muscle stem cells, expanded them, formed tissue, and implanted it under the skin of aged animals. The resulting bio‑transplants (myografts) autonomously built a vascular network and began contracting spontaneously 24 hours a day, 7 days a week, without any nervous‑system or brain input.
The contracting muscle acts as a continuous biological factory, releasing myokines and signaling molecules into the bloodstream. Old mice receiving these subcutaneous “patches” showed increased lean body mass, stronger grip, better treadmill endurance, higher bone density, and reduced inflammation markers.
In the mice brains, the number of degrading neurons in the hippocampus fell, BDNF levels rose, and spatial memory improved. (Although the brain‑test sample was tiny – only 3 individuals per group.)
The myograft is not just a gym mimic but a removable biological reactor. Researchers genetically engineered the implanted cells to secrete parathyroid hormone (PTH) and growth hormone, giving a stable blood protein level without the spikes and drops seen with injections.
The experiments used Matrigel matrix, which is unsuitable for humans, so clinical translation will require a different scaffold and scaling of autologous cell production. Nevertheless, the concept of a subcutaneous muscle‑doser has already proven viable.
🔗 Read original →
Scientists took mouse muscle stem cells, expanded them, formed tissue, and implanted it under the skin of aged animals. The resulting bio‑transplants (myografts) autonomously built a vascular network and began contracting spontaneously 24 hours a day, 7 days a week, without any nervous‑system or brain input.
The contracting muscle acts as a continuous biological factory, releasing myokines and signaling molecules into the bloodstream. Old mice receiving these subcutaneous “patches” showed increased lean body mass, stronger grip, better treadmill endurance, higher bone density, and reduced inflammation markers.
In the mice brains, the number of degrading neurons in the hippocampus fell, BDNF levels rose, and spatial memory improved. (Although the brain‑test sample was tiny – only 3 individuals per group.)
The myograft is not just a gym mimic but a removable biological reactor. Researchers genetically engineered the implanted cells to secrete parathyroid hormone (PTH) and growth hormone, giving a stable blood protein level without the spikes and drops seen with injections.
The experiments used Matrigel matrix, which is unsuitable for humans, so clinical translation will require a different scaffold and scaling of autologous cell production. Nevertheless, the concept of a subcutaneous muscle‑doser has already proven viable.
🔗 Read original →
Nature
Contractile myografts confer systemic anti-aging benefits
Nature Aging - Cell and gene therapy are of interest to treat age-related diseases. Here the authors developed subcutaneous transplantation of autologous myocytes to form myografts. Myografts...
Actin cytoskeleton disruption shortens worm lifespan, while mild stabilization extends it
On August 24, a bibliographic record of an iScience article appeared in Crossref. In the full text, authors altered actin cytoskeleton function in the roundworm Caenorhabditis elegans: network disruption shortened life, while mild chemical stabilization extended it; a high dose produced the opposite effect.
The actin cytoskeleton helps muscles contract, maintains intestinal cell junctions, and participates in intracellular transport. In prior work from the same research line, increased production of the BET‑1 protein preserved actin in old worms and extended their lifespan.
The new study examines the converse: what happens when the network loses order. Authors sequentially weakened actin and three proteins that direct its assembly, disassembly, and anchoring. In muscle, intestine, and cuticle, filaments lost normal organization earlier, and mobility declined with age. In some worms, gene activity resembled that of older animals.
Because the age of intervention matters — in another C. elegans experiment, transient NuA4 suppression in early development extended lifespan, while later intervention shortened it — part of the new experiments began on the first day of adult life. Lifespan reduction persisted upon suppression of arx-2, a component of the Arp2/3 complex that builds branched actin networks.
Separately, adult worms were given two substances. Latrunculin A breaks actin filaments and, with increasing dose, shortened life. Jasplakinolide helps filaments assemble and stabilize: at low concentrations it extended life, at high concentrations it shortened. Authors link the harm of high doses to overly stabilized filaments being less able to remodel.
Actin disruption simultaneously altered mitochondrial shape and function, protein homeostasis, cellular component recycling, and intestinal barrier integrity. In two independent human cohorts, authors also correlated ACTB gene variants with the pace of age‑related walking slowdown.
🔗 Read original →
On August 24, a bibliographic record of an iScience article appeared in Crossref. In the full text, authors altered actin cytoskeleton function in the roundworm Caenorhabditis elegans: network disruption shortened life, while mild chemical stabilization extended it; a high dose produced the opposite effect.
The actin cytoskeleton helps muscles contract, maintains intestinal cell junctions, and participates in intracellular transport. In prior work from the same research line, increased production of the BET‑1 protein preserved actin in old worms and extended their lifespan.
The new study examines the converse: what happens when the network loses order. Authors sequentially weakened actin and three proteins that direct its assembly, disassembly, and anchoring. In muscle, intestine, and cuticle, filaments lost normal organization earlier, and mobility declined with age. In some worms, gene activity resembled that of older animals.
Because the age of intervention matters — in another C. elegans experiment, transient NuA4 suppression in early development extended lifespan, while later intervention shortened it — part of the new experiments began on the first day of adult life. Lifespan reduction persisted upon suppression of arx-2, a component of the Arp2/3 complex that builds branched actin networks.
Separately, adult worms were given two substances. Latrunculin A breaks actin filaments and, with increasing dose, shortened life. Jasplakinolide helps filaments assemble and stabilize: at low concentrations it extended life, at high concentrations it shortened. Authors link the harm of high doses to overly stabilized filaments being less able to remodel.
Actin disruption simultaneously altered mitochondrial shape and function, protein homeostasis, cellular component recycling, and intestinal barrier integrity. In two independent human cohorts, authors also correlated ACTB gene variants with the pace of age‑related walking slowdown.
🔗 Read original →
iScience
Form and function of actin impacts actin health and aging
Cell biology; Bioinformatics; Omics
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). There's no explicit citation in the text; but maybe we can treat "BixBench3" as a benchmark? Not a citation. The text mentions "26 августа компания Edison Scientific представила BixBench3". That's a date. So we can wrap the date in double asterisks? It's a date, yes. Also numbers: 20 tasks, 13 models, best average result 48%, 138 checkable results, 4-14 results per task, best model GPT 5.6 Sol average 48%, tasks with data >100 GB average 10%, 50-100 GB average 34%, <50 GB average 37%, when result depended on three or more sequential steps average 24%. That's many numbers; we need to select at most 4-5 important facts to bold. Choose: date (August 26), 20 tasks, 13 models, best average 48%, >100 GB 10%, 50-100 GB 34%, <50 GB 37%, three+ steps 24%. That's many. We need max 4-5. Let's pick: date, number of tasks (20), best model average (48%), performance on >100 GB (10%), and performance on three+ sequential steps (24%). That's 5.
We'll bold those: August 26, 20 tasks, 48%, 10%, 24%. Ensure not whole sentence bolded.
Now we need to wrap study/journal citations in single underscores. There's none; maybe we can treat "BixBench3" as a benchmark but not a citation. Could we add a placeholder? No, we must not add invented detail. So we leave none.
Now produce English translation.
Let's craft headline: "Edison Scientific launches BixBench3 to test AI agents on reproducing biological analysis". Count characters: Let's count: "Edison Scientific launches BixBench3 to test AI agents on reproducing biological analysis". Count: Edison(6) + space=1 =>7, Scientific(11)=>18, space=19, launches(8)=>27, space=28, BixBench3(9)=>37, space=38, to(2)=>40, space=41, test(4)=>45, space=46, AI(2)=>48, space=49, agents(6)=>55, space=56, on(2)=>58, space=59, reproducing(11)=>70, space=71, biological(10)=>81, space=82, analysis(8)=>90. Exactly 90 characters? Let's recount precisely: "Edison Scientific launches BixBench3 to test AI agents on reproducing biological analysis"
Characters:
E(1)
d2
i3
s4
o5
n6
space7
S8
c9
i10
e11
n12
t13
i14
f15
i16
c17
space18
l19
a20
u21
n22
c23
h24
e25
s26
space27
B28
i29
x30
B31
e32
n33
c34
h35
3 36
space37
t38
o39
space40
t41
e42
s43
t44
space45
A46
I47
space48
a49
g50
e51
n52
t53
s54
space55
o56
n57
space58
r59
e60
p61
r62
o63
d64
u65
c66
i67
n68
g69
space70
b71
i72
o73
l74
o75
g76
i77
c78
a79
l80
space81
a82
n83
a84
l85
y86
s87
i88
s89
.90
So 90 characters including period. Under 90 chars? It says under 90 chars. This is exactly 90, maybe need less than 90. Remove period. Then 89 characters. So headline without period: "Edison Scientific launches BixBench3 to test AI agents on reproducing biological analysis". Count without period: 89. Good.
Now body: need short paragraphs 2-3 sentences each.
We'll translate content.
Paragraph 1: Introduce BixBench3 release date, number of tasks, description.
Paragraph 2: Explain process: agent gets goal, methods, raw data, produce intermediate results, compare to original work, early error propagates.
Paragraph 3: Details of benchmark: 20 papers, 138 checkable results, each task yields 4-14 results, scoring based on preserving biological interpretation.…
🔗 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). There's no explicit citation in the text; but maybe we can treat "BixBench3" as a benchmark? Not a citation. The text mentions "26 августа компания Edison Scientific представила BixBench3". That's a date. So we can wrap the date in double asterisks? It's a date, yes. Also numbers: 20 tasks, 13 models, best average result 48%, 138 checkable results, 4-14 results per task, best model GPT 5.6 Sol average 48%, tasks with data >100 GB average 10%, 50-100 GB average 34%, <50 GB average 37%, when result depended on three or more sequential steps average 24%. That's many numbers; we need to select at most 4-5 important facts to bold. Choose: date (August 26), 20 tasks, 13 models, best average 48%, >100 GB 10%, 50-100 GB 34%, <50 GB 37%, three+ steps 24%. That's many. We need max 4-5. Let's pick: date, number of tasks (20), best model average (48%), performance on >100 GB (10%), and performance on three+ sequential steps (24%). That's 5.
We'll bold those: August 26, 20 tasks, 48%, 10%, 24%. Ensure not whole sentence bolded.
Now we need to wrap study/journal citations in single underscores. There's none; maybe we can treat "BixBench3" as a benchmark but not a citation. Could we add a placeholder? No, we must not add invented detail. So we leave none.
Now produce English translation.
Let's craft headline: "Edison Scientific launches BixBench3 to test AI agents on reproducing biological analysis". Count characters: Let's count: "Edison Scientific launches BixBench3 to test AI agents on reproducing biological analysis". Count: Edison(6) + space=1 =>7, Scientific(11)=>18, space=19, launches(8)=>27, space=28, BixBench3(9)=>37, space=38, to(2)=>40, space=41, test(4)=>45, space=46, AI(2)=>48, space=49, agents(6)=>55, space=56, on(2)=>58, space=59, reproducing(11)=>70, space=71, biological(10)=>81, space=82, analysis(8)=>90. Exactly 90 characters? Let's recount precisely: "Edison Scientific launches BixBench3 to test AI agents on reproducing biological analysis"
Characters:
E(1)
d2
i3
s4
o5
n6
space7
S8
c9
i10
e11
n12
t13
i14
f15
i16
c17
space18
l19
a20
u21
n22
c23
h24
e25
s26
space27
B28
i29
x30
B31
e32
n33
c34
h35
3 36
space37
t38
o39
space40
t41
e42
s43
t44
space45
A46
I47
space48
a49
g50
e51
n52
t53
s54
space55
o56
n57
space58
r59
e60
p61
r62
o63
d64
u65
c66
i67
n68
g69
space70
b71
i72
o73
l74
o75
g76
i77
c78
a79
l80
space81
a82
n83
a84
l85
y86
s87
i88
s89
.90
So 90 characters including period. Under 90 chars? It says under 90 chars. This is exactly 90, maybe need less than 90. Remove period. Then 89 characters. So headline without period: "Edison Scientific launches BixBench3 to test AI agents on reproducing biological analysis". Count without period: 89. Good.
Now body: need short paragraphs 2-3 sentences each.
We'll translate content.
Paragraph 1: Introduce BixBench3 release date, number of tasks, description.
Paragraph 2: Explain process: agent gets goal, methods, raw data, produce intermediate results, compare to original work, early error propagates.
Paragraph 3: Details of benchmark: 20 papers, 138 checkable results, each task yields 4-14 results, scoring based on preserving biological interpretation.…
🔗 Read original →
We Will Cure Releases Longevity Biotech Clinical Trials Map Through 2027
On August 25, the editorial project We Will Cure published a map of longevity biotech clinical trials planned through 2027. It aligns programs that have already begun human studies, obtained initial data, or are preparing the next stage. By the end of 2027 they will answer whether humans tolerate the intervention, whether it affects the targeted process, and whether there is grounds to expand the trial.
Companies test aging‑biology ideas through concrete diseases and quantifiable
🔗 Read original →
On August 25, the editorial project We Will Cure published a map of longevity biotech clinical trials planned through 2027. It aligns programs that have already begun human studies, obtained initial data, or are preparing the next stage. By the end of 2027 they will answer whether humans tolerate the intervention, whether it affects the targeted process, and whether there is grounds to expand the trial.
Companies test aging‑biology ideas through concrete diseases and quantifiable
🔗 Read original →
PubMed Central (PMC)
Recommendations for biomarker data collection in clinical trials by longevity biotechnology companies
Biomarkers of aging have the potential to transform geroscience clinical trials because of their broad applications in stratifying participants, prioritizing interventions, and monitoring responses to geroprotectors. As longevity biotechnology ...
Cryoprotectant penetrates fixed human brain over nine months
On 24 August PLOS One published a protocol for storing whole human brains at −20 °C after chemically fixing their cellular structure. The goal was to assess how a cryoprotectant solution diffuses through the tissue and whether fine structure survives cooling, storage, and rewarming.
Brains were first fixed with a cross‑linking agent to lock in tissue architecture after death.
🔗 Read original →
On 24 August PLOS One published a protocol for storing whole human brains at −20 °C after chemically fixing their cellular structure. The goal was to assess how a cryoprotectant solution diffuses through the tissue and whether fine structure survives cooling, storage, and rewarming.
Brains were first fixed with a cross‑linking agent to lock in tissue architecture after death.
🔗 Read original →
PubMed Central (PMC)
Cryopreservation of brain cell structure: a review
Cryopreservation, the preservation of tissues at subzero temperatures, is a mainstay of brain banking that allows for the storage of brain tissue without the use of chemical fixatives. This is particularly important for molecular studies that are ...