We need to translate Russian news into English, format with headline line under 90 chars, blank line, then short paragraphs 2-3 sentences each, separated by blank lines. Wrap few 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 (e.g., Nature Aging, July 2026). There are no explicit citations in the Russian text; we should not add any. So we will not use underscores unless there is a citation; none present. So we will not use underscores.
We need to select at most 4-5 double-asterisk facts. Important facts: 9 billion possible SNVs, 27,000 predictions per mutation, 100 million indels, 2,500 motifs, AVI score definition, 98% noncoding, 2% coding, test on patients, narrowing from 526 to 4 candidates in UK Biobank of 54,000 people. Also mention DeepMind, AlphaGenome Atlas, free for academics, commercial via Google Cloud soon. Need to pick maybe 5 facts: 9 billion SNVs, 27k predictions per mutation, 100 million indels, AVI score definition (AVI 10 top 10%, AVI 30 top 0.1%), 98% noncoding, 2% coding (that's two numbers but could count as one fact? maybe combine). We need at most 4-5; we can choose 5: 9 billion SNVs, 27k predictions per mutation, 100 million indels, AVI score definition (maybe two subpoints but could be considered one fact), 98% noncoding / 2% coding (maybe considered one). That's 5.
We must wrap each fact in double asterisks. Ensure we don't wrap whole sentences. So embed numbers etc.
We need to keep the translation natural. Let's craft.
Headline: maybe "DeepMind launches AlphaGenome Atlas, AI‑driven catalog of 9 billion human SNVs". Under 90 chars.
Then blank line.
Paragraph 1: describe database, free for academics, commercial via Google Cloud soon.
Paragraph 2: how built: AlphaGenome network analyzed reference genome, each base substitution yields predictions, each SNV has avg 27k predictions across cell types, etc.
Paragraph 3: includes indels, motifs, noncoding vs coding insight.
Paragraph 4: AVI score explanation.
Paragraph 5: real‑patient test results, narrowing candidates.
Make sure each paragraph 2-3 sentences.
Now embed double asterisks around selected facts.
Facts:
- 9 billion possible single‑nucleotide mutations
- each SNV accompanied by ≈27 000 individual predictions
- >100 million insertions and deletions
- AVI score: AVI 10 = top 10 % impact, AVI 30 = top 0.1 % impact
- 98 % of the genome is non‑coding (regulatory), only 2 % codes for proteins
We need to ensure we don't exceed 4-5 facts; we have 5 facts (the last includes two percentages but it's one fact about composition). Good.
Now ensure we don't accidentally double asterisk other numbers like 54 000, 526, 4, etc. Those are not to be highlighted per rule (we can leave them plain). Also avoid wrapping whole sentences.
Now produce final answer.
Check length of headline: "DeepMind launches AlphaGenome Atlas, AI‑driven catalog of 9 billion human SNVs". Count characters: Let's count roughly: DeepMind (8) + space (1) =9, launches (8) =>17, space=18, AlphaGenome (11) =>29, space=30, Atlas, (6 inc comma?) Actually "Atlas," 6 inc comma =>36, space=37, AI‑driven (9 inc hyphen?) =>46, space=47, catalog (7)=>54, space=55, of (2)=>57, space=58, 9 billion (maybe 10 inc space?) Actually "9 billion" includes a space? We have space before 9? Already have space after of. So "9 billion" is 10 inc space? Let's just trust it's under 90. It's fine.
Now paragraphs.
Paragraph 1: "DeepMind has announced the release of the AlphaGenome Atlas, a free scientific catalogue that stores AI‑generated predictions of the biological consequences of every possible single‑nucleotide change in the human genome. Academic researchers can access the database at no cost, while commercial access via Google Cloud is slated to launch in the near future."…
🔗 Read original →
We need to select at most 4-5 double-asterisk facts. Important facts: 9 billion possible SNVs, 27,000 predictions per mutation, 100 million indels, 2,500 motifs, AVI score definition, 98% noncoding, 2% coding, test on patients, narrowing from 526 to 4 candidates in UK Biobank of 54,000 people. Also mention DeepMind, AlphaGenome Atlas, free for academics, commercial via Google Cloud soon. Need to pick maybe 5 facts: 9 billion SNVs, 27k predictions per mutation, 100 million indels, AVI score definition (AVI 10 top 10%, AVI 30 top 0.1%), 98% noncoding, 2% coding (that's two numbers but could count as one fact? maybe combine). We need at most 4-5; we can choose 5: 9 billion SNVs, 27k predictions per mutation, 100 million indels, AVI score definition (maybe two subpoints but could be considered one fact), 98% noncoding / 2% coding (maybe considered one). That's 5.
We must wrap each fact in double asterisks. Ensure we don't wrap whole sentences. So embed numbers etc.
We need to keep the translation natural. Let's craft.
Headline: maybe "DeepMind launches AlphaGenome Atlas, AI‑driven catalog of 9 billion human SNVs". Under 90 chars.
Then blank line.
Paragraph 1: describe database, free for academics, commercial via Google Cloud soon.
Paragraph 2: how built: AlphaGenome network analyzed reference genome, each base substitution yields predictions, each SNV has avg 27k predictions across cell types, etc.
Paragraph 3: includes indels, motifs, noncoding vs coding insight.
Paragraph 4: AVI score explanation.
Paragraph 5: real‑patient test results, narrowing candidates.
Make sure each paragraph 2-3 sentences.
Now embed double asterisks around selected facts.
Facts:
- 9 billion possible single‑nucleotide mutations
- each SNV accompanied by ≈27 000 individual predictions
- >100 million insertions and deletions
- AVI score: AVI 10 = top 10 % impact, AVI 30 = top 0.1 % impact
- 98 % of the genome is non‑coding (regulatory), only 2 % codes for proteins
We need to ensure we don't exceed 4-5 facts; we have 5 facts (the last includes two percentages but it's one fact about composition). Good.
Now ensure we don't accidentally double asterisk other numbers like 54 000, 526, 4, etc. Those are not to be highlighted per rule (we can leave them plain). Also avoid wrapping whole sentences.
Now produce final answer.
Check length of headline: "DeepMind launches AlphaGenome Atlas, AI‑driven catalog of 9 billion human SNVs". Count characters: Let's count roughly: DeepMind (8) + space (1) =9, launches (8) =>17, space=18, AlphaGenome (11) =>29, space=30, Atlas, (6 inc comma?) Actually "Atlas," 6 inc comma =>36, space=37, AI‑driven (9 inc hyphen?) =>46, space=47, catalog (7)=>54, space=55, of (2)=>57, space=58, 9 billion (maybe 10 inc space?) Actually "9 billion" includes a space? We have space before 9? Already have space after of. So "9 billion" is 10 inc space? Let's just trust it's under 90. It's fine.
Now paragraphs.
Paragraph 1: "DeepMind has announced the release of the AlphaGenome Atlas, a free scientific catalogue that stores AI‑generated predictions of the biological consequences of every possible single‑nucleotide change in the human genome. Academic researchers can access the database at no cost, while commercial access via Google Cloud is slated to launch in the near future."…
🔗 Read original →
Google
AlphaGenome
AlphaGenome – Access Google DeepMind’s unifying genomics model for deciphering DNA function.
Heart MRI framing and processing shape genetic discovery in UK Biobank study
On September 4 Devanj Pandey and Wagish Narasimhan posted a preprint analyzing cardiac MRI from 69,932 participants in the UK Biobank. Their self‑supervised model reconstructed masked frames of the cine‑MRI videos without physician labels and used the resulting numerical features to search for DNA‑variant associations.
Cine‑MRI shows a beating heart that occupies only about one‑quarter to one‑third of each frame, with the rest filled by chest, spine and acquisition details. The model turned each participant’s three cine‑MRI projections into a set of 2,304 numbers, which were then compressed to the 20 principal components representing the largest differences between individuals.
For predicting the scan site the model achieved R² = 0.55, and for predicting left‑ventricular ejection fraction it reached R² = 0.37; higher R² indicates more accurate prediction. These values show how much of the variance in each trait is captured by the imaging‑derived features.
Before the genome‑wide search the researchers cropped the frames to the moving heart region, removing roughly 65–75% of the chest field, and regressed out age, sex, body‑mass index, height and scan site from all 2,304 features. Only after this cleanup were the 20 principal components recomputed for the genetic analysis.
Using the full feature set they identified 101 genome loci, including all 11 previously known heart‑related regions. A later adjustment that corrected the principal components yielded 60 loci and one novel candidate. In a hold‑out sample of 13,717 participants whose scans were not used for model training, the effect sizes of the top variants matched those from the full sample.
🔗 Read original →
On September 4 Devanj Pandey and Wagish Narasimhan posted a preprint analyzing cardiac MRI from 69,932 participants in the UK Biobank. Their self‑supervised model reconstructed masked frames of the cine‑MRI videos without physician labels and used the resulting numerical features to search for DNA‑variant associations.
Cine‑MRI shows a beating heart that occupies only about one‑quarter to one‑third of each frame, with the rest filled by chest, spine and acquisition details. The model turned each participant’s three cine‑MRI projections into a set of 2,304 numbers, which were then compressed to the 20 principal components representing the largest differences between individuals.
For predicting the scan site the model achieved R² = 0.55, and for predicting left‑ventricular ejection fraction it reached R² = 0.37; higher R² indicates more accurate prediction. These values show how much of the variance in each trait is captured by the imaging‑derived features.
Before the genome‑wide search the researchers cropped the frames to the moving heart region, removing roughly 65–75% of the chest field, and regressed out age, sex, body‑mass index, height and scan site from all 2,304 features. Only after this cleanup were the 20 principal components recomputed for the genetic analysis.
Using the full feature set they identified 101 genome loci, including all 11 previously known heart‑related regions. A later adjustment that corrected the principal components yielded 60 loci and one novel candidate. In a hold‑out sample of 13,717 participants whose scans were not used for model training, the effect sizes of the top variants matched those from the full sample.
🔗 Read original →
medRxiv
Field-of-view confounding shapes genetic discovery from self-supervised cardiac-imaging phenotypes
Self-supervised models increasingly convert medical images into quantitative phenotypes for biological discovery, but statistical reproducibility does not establish that a learned phenotype represents the intended anatomy. We trained a video masked-autoencoder…
John Tower proposes model linking adaptation, aging, and sexual conflict
On September 7, biologist John Tower of the University of Southern California posted a preprint on bioRxiv outlining a model that connects adaptive state switching with aging and sexual conflict. He describes how a regulated exit from a maintained state can be beneficial now but costly later, with cells needing to preserve functional configurations while sometimes requiring a different response
🔗 Read original →
On September 7, biologist John Tower of the University of Southern California posted a preprint on bioRxiv outlining a model that connects adaptive state switching with aging and sexual conflict. He describes how a regulated exit from a maintained state can be beneficial now but costly later, with cells needing to preserve functional configurations while sometimes requiring a different response
🔗 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 leading '#'.
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:
- In developing mouse heart, 23 of 50 tested genes altered cardiomyocyte maturation programs.
- September 7 2026 article in Nature Cardiovascular Research about PIP-seq method linking gene knockout to cardiomyocyte RNA changes.
- Pilot screen: 23 genes altered at least one maturation program; 21 of them were previously uncharacterized regulators.
- After birth, cardiomyocytes grow, organize sarcomeres, change energy metabolism and electrical properties.
- Authors tracked transition at birth and at 7, 14, 21 days: read RNA of 67 thousand nuclei and mapped position of 402 thousand cells on heart sections.
- Largest gene activity changes occurred in first 14 days; therefore knockout effect evaluated at day 14.
- To test candidate function, injected AAV9 library into hearts of newborn mice; each construct carried two CRISPR guide RNAs leading Cas9 to target gene.
- After tissue fixation, PIP-seq simultaneously identified in nucleus the guide, editing marker of its gene, and RNA produced after intervention. So one experiment linked specific gene knockout to cardiomyocyte response within heart.
- Screen yielded 5,706 nuclei with assigned guide.
- Genes Srf and Ppargc1a, whose role in maturation already known, served as positive controls and gave expected maturation program disruptions.
- Knocking out Rad21, a gene linked to DNA packaging, altered activity of genes related to contractile fiber assembly, electrical properties, and fatty acid metabolism.
- Authors separately knocked out gene Rreb1.
- By day 35 after birth, cardiomyocytes with Rreb1 knockout were shorter, and their T-tubules — membrane channels conducting electrical signal into cell — more often had disrupted structure.
- Mitochondria of these cells consumed less oxygen at maximal load.
- Mice with Rreb1 knockout showed reduced fractional shortening and ejection fraction — two measures of cardiac contractile function on ultrasound.
- Separate validation linked screen finding to changes in cell structure, its energetic work, and heart function.
We need to wrap key numbers, percentages, drug/company/gene names, dates in double asterisks. At most 4-5 per post.
Select important facts: numbers: 23 of 50 genes, September 7 2026, 67 thousand nuclei, 402 thousand cells, first 14 days, day 14 evaluation, 5,706 nuclei, day 35, fractional shortening/ejection fraction (maybe not numeric). Gene names: Srf, Ppargc1a, Rad21, Rreb1. Also AAV9. Also PIP-seq. Also Nature Cardiovascular Research.
We need to limit double asterisks to at most 4-5 per post. So we need to be selective: maybe highlight: 23 of 50 genes, September 7 2026, 67 000 nuclei, 402 000 cells, day 14, 5 706 nuclei, day 35, Srf, Ppargc1a, Rad21, Rreb1... That's many. We need only 4-5 total. So we must choose only a few key facts.
Which are most important? Probably the screen results: 23 of 50 genes altered; 21 previously uncharacterized; the date; the number of nuclei/cells; the day 14 evaluation; the day 35 phenotype; maybe the gene names of interest: Rreb1.
But we need to keep double asterisks limited. Let's decide to wrap: 23 of 50 genes, September 7 2026, 67 000 nuclei, 402 000 cells, day 14 (or day 35). That's 5.…
🔗 Read original →
First line: short headline under 90 chars, no markdown, no leading '#'.
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:
- In developing mouse heart, 23 of 50 tested genes altered cardiomyocyte maturation programs.
- September 7 2026 article in Nature Cardiovascular Research about PIP-seq method linking gene knockout to cardiomyocyte RNA changes.
- Pilot screen: 23 genes altered at least one maturation program; 21 of them were previously uncharacterized regulators.
- After birth, cardiomyocytes grow, organize sarcomeres, change energy metabolism and electrical properties.
- Authors tracked transition at birth and at 7, 14, 21 days: read RNA of 67 thousand nuclei and mapped position of 402 thousand cells on heart sections.
- Largest gene activity changes occurred in first 14 days; therefore knockout effect evaluated at day 14.
- To test candidate function, injected AAV9 library into hearts of newborn mice; each construct carried two CRISPR guide RNAs leading Cas9 to target gene.
- After tissue fixation, PIP-seq simultaneously identified in nucleus the guide, editing marker of its gene, and RNA produced after intervention. So one experiment linked specific gene knockout to cardiomyocyte response within heart.
- Screen yielded 5,706 nuclei with assigned guide.
- Genes Srf and Ppargc1a, whose role in maturation already known, served as positive controls and gave expected maturation program disruptions.
- Knocking out Rad21, a gene linked to DNA packaging, altered activity of genes related to contractile fiber assembly, electrical properties, and fatty acid metabolism.
- Authors separately knocked out gene Rreb1.
- By day 35 after birth, cardiomyocytes with Rreb1 knockout were shorter, and their T-tubules — membrane channels conducting electrical signal into cell — more often had disrupted structure.
- Mitochondria of these cells consumed less oxygen at maximal load.
- Mice with Rreb1 knockout showed reduced fractional shortening and ejection fraction — two measures of cardiac contractile function on ultrasound.
- Separate validation linked screen finding to changes in cell structure, its energetic work, and heart function.
We need to wrap key numbers, percentages, drug/company/gene names, dates in double asterisks. At most 4-5 per post.
Select important facts: numbers: 23 of 50 genes, September 7 2026, 67 thousand nuclei, 402 thousand cells, first 14 days, day 14 evaluation, 5,706 nuclei, day 35, fractional shortening/ejection fraction (maybe not numeric). Gene names: Srf, Ppargc1a, Rad21, Rreb1. Also AAV9. Also PIP-seq. Also Nature Cardiovascular Research.
We need to limit double asterisks to at most 4-5 per post. So we need to be selective: maybe highlight: 23 of 50 genes, September 7 2026, 67 000 nuclei, 402 000 cells, day 14, 5 706 nuclei, day 35, Srf, Ppargc1a, Rad21, Rreb1... That's many. We need only 4-5 total. So we must choose only a few key facts.
Which are most important? Probably the screen results: 23 of 50 genes altered; 21 previously uncharacterized; the date; the number of nuclei/cells; the day 14 evaluation; the day 35 phenotype; maybe the gene names of interest: Rreb1.
But we need to keep double asterisks limited. Let's decide to wrap: 23 of 50 genes, September 7 2026, 67 000 nuclei, 402 000 cells, day 14 (or day 35). That's 5.…
🔗 Read original →
Nature
Spatially guided in vivo single-cell functional genomics of postnatal heart
Nature Cardiovascular Research - By integrating single-nucleus RNA sequencing and spatial transcriptomics, Wang, Dong, Song et al. generated a high-resolution spatiotemporal atlas of the postnatal...
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.
We need to identify citations: The text mentions "технической работе команды" but no explicit journal citation. It mentions "В данных Британского биобанка UK Biobank — 54 189 участников и уровни 2 028 белков в крови". No citation.
Thus we may have no citations to wrap. But we can still wrap any mention like "UK Biobank" maybe not needed. The rule: wrap study/journal citations and publication references in single underscores. If none, we just don't use underscores.
We need to select at most 4-5 important facts to wrap in . Likely: 9 billion possible single-letter DNA substitutions, date September 8, AVI score, DNM1 gene, 12 intron variants, 3 previously described pathogenic variants, 54,189 participants, 2,028 blood proteins, 728 links vs 595 baseline, 22% increase.
We need to be selective: at most 4-5 per post. Choose maybe: 9 billion, September 8, DNM1 gene, 12 intron variants (3 pathogenic), 54,189 participants, 2,028 proteins, 728 links, 22% increase. That's many. Need limit to 4-5.
We can combine some: maybe wrap "9 billion", "8 September", "DNM1", "54 189 participants", "22 % increase". That's 5.
Alternatively include "AVI score". But we need to be selective. Let's decide: 9 billion, 8 September, DNM1, 54 189 participants, 22 % increase. That's five.
We must ensure we don't wrap whole sentence. Just wrap those tokens.
Now produce translation.
First line headline: maybe "DeepMind releases AlphaGenome Atlas mapping 9 billion DNA single‑letter changes". Count chars: Let's count: "DeepMind releases AlphaGenome Atlas mapping 9 billion DNA single‑letter changes". Count characters including spaces: DeepMind(8) + space=1 =>9, releases(8)=>17, space=18, AlphaGenome(11)=>29, space=30, Atlas(5)=>35, space=36, mapping(6)=>42, space=43, 9 billion (maybe "9 billion" counts as 2? "9" plus space? Actually "9 billion": "9" (1) + non-breaking space? We'll count as "9" (1) + space (1) + "billion"(7)=9 => total 43+9=52, space=53, DNA(3)=56, space=57, single‑letter (maybe hyphen counts) "single‑letter": 13? Let's approximate: s(1)i2n3g4l5e6‑7l8e9t10t11e12r13 =>13, space => after that? Actually "single‑letter" then space then "changes". So add 13 =>70, space=71, changes(7)=78. So under 90. Good.
Now blank line.
Now body paragraphs.
We need to split into short paragraphs 2-3 sentences each.
Let's craft translation:
Paragraph 1: Introduce AlphaGenome Atlas, previous model, new map.
Paragraph 2: Explain purpose, AVI score, components.
Paragraph 3: Example with DNM1 gene, splicing, experimental validation.
Paragraph 4: UK Biobank analysis, grouping rare non-coding variants, results.
Paragraph 5: Note that model input is DNA sequence, need experimental verification.
Make sure each paragraph 2-3 sentences.
Now incorporate double asterisks around selected facts.
We need to ensure we don't accidentally wrap whole sentence.
We'll embed 9 billion, 8 September, DNM1, 54 189 participants, 22 % increase.
Check each appears only once.
Now produce final.
Let's draft:…
🔗 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.
We need to identify citations: The text mentions "технической работе команды" but no explicit journal citation. It mentions "В данных Британского биобанка UK Biobank — 54 189 участников и уровни 2 028 белков в крови". No citation.
Thus we may have no citations to wrap. But we can still wrap any mention like "UK Biobank" maybe not needed. The rule: wrap study/journal citations and publication references in single underscores. If none, we just don't use underscores.
We need to select at most 4-5 important facts to wrap in . Likely: 9 billion possible single-letter DNA substitutions, date September 8, AVI score, DNM1 gene, 12 intron variants, 3 previously described pathogenic variants, 54,189 participants, 2,028 blood proteins, 728 links vs 595 baseline, 22% increase.
We need to be selective: at most 4-5 per post. Choose maybe: 9 billion, September 8, DNM1 gene, 12 intron variants (3 pathogenic), 54,189 participants, 2,028 proteins, 728 links, 22% increase. That's many. Need limit to 4-5.
We can combine some: maybe wrap "9 billion", "8 September", "DNM1", "54 189 participants", "22 % increase". That's 5.
Alternatively include "AVI score". But we need to be selective. Let's decide: 9 billion, 8 September, DNM1, 54 189 participants, 22 % increase. That's five.
We must ensure we don't wrap whole sentence. Just wrap those tokens.
Now produce translation.
First line headline: maybe "DeepMind releases AlphaGenome Atlas mapping 9 billion DNA single‑letter changes". Count chars: Let's count: "DeepMind releases AlphaGenome Atlas mapping 9 billion DNA single‑letter changes". Count characters including spaces: DeepMind(8) + space=1 =>9, releases(8)=>17, space=18, AlphaGenome(11)=>29, space=30, Atlas(5)=>35, space=36, mapping(6)=>42, space=43, 9 billion (maybe "9 billion" counts as 2? "9" plus space? Actually "9 billion": "9" (1) + non-breaking space? We'll count as "9" (1) + space (1) + "billion"(7)=9 => total 43+9=52, space=53, DNA(3)=56, space=57, single‑letter (maybe hyphen counts) "single‑letter": 13? Let's approximate: s(1)i2n3g4l5e6‑7l8e9t10t11e12r13 =>13, space => after that? Actually "single‑letter" then space then "changes". So add 13 =>70, space=71, changes(7)=78. So under 90. Good.
Now blank line.
Now body paragraphs.
We need to split into short paragraphs 2-3 sentences each.
Let's craft translation:
Paragraph 1: Introduce AlphaGenome Atlas, previous model, new map.
Paragraph 2: Explain purpose, AVI score, components.
Paragraph 3: Example with DNM1 gene, splicing, experimental validation.
Paragraph 4: UK Biobank analysis, grouping rare non-coding variants, results.
Paragraph 5: Note that model input is DNA sequence, need experimental verification.
Make sure each paragraph 2-3 sentences.
Now incorporate double asterisks around selected facts.
We need to ensure we don't accidentally wrap whole sentence.
We'll embed 9 billion, 8 September, DNM1, 54 189 participants, 22 % increase.
Check each appears only once.
Now produce final.
Let's draft:…
🔗 Read original →
Nature
Advancing regulatory variant effect prediction with AlphaGenome
Nature - AlphaGenome, a deep learning model that inputs 1-Mb DNA sequence to predict functional genomic tracks at single-base resolution across diverse modalities, outperforms existing models in...
MetaGNN authors retract claim that reaction links boosted model accuracy
In a September 7 preprint, the authors of MetaGNN retracted their earlier claim that incorporating reaction‑network links improved model performance by 0.105 AUROC. They said the claim was based on a flawed benchmark.
The model was evaluated on 5,925 reactions labeled ‘active’ by applying a threshold to the input gene‑activity scores. Simply ranking those scores gives an AUROC of 1.0000, indicating the benchmark only tested whether the model could reproduce the rule used to create the labels.
For a separate set of 4,675 reactions, the labels could be reproduced exactly from a fixed RandomState(42) generator, and the same label set appeared for every patient. In colon‑cancer analyses, the input features for 624 patients were all zeros, so the model had no patient‑specific variation to learn from.
After reconstructing the input data, MetaGNN obtained an AUROC of 0.5800. Ranking the raw gene‑expression values yielded 0.6342, while a baseline that only records whether a gene is expressed gave 0.608
🔗 Read original →
In a September 7 preprint, the authors of MetaGNN retracted their earlier claim that incorporating reaction‑network links improved model performance by 0.105 AUROC. They said the claim was based on a flawed benchmark.
The model was evaluated on 5,925 reactions labeled ‘active’ by applying a threshold to the input gene‑activity scores. Simply ranking those scores gives an AUROC of 1.0000, indicating the benchmark only tested whether the model could reproduce the rule used to create the labels.
For a separate set of 4,675 reactions, the labels could be reproduced exactly from a fixed RandomState(42) generator, and the same label set appeared for every patient. In colon‑cancer analyses, the input features for 624 patients were all zeros, so the model had no patient‑specific variation to learn from.
After reconstructing the input data, MetaGNN obtained an AUROC of 0.5800. Ranking the raw gene‑expression values yielded 0.6342, while a baseline that only records whether a gene is expressed gave 0.608
🔗 Read original →
bioRxiv
Benchmark validity in graph neural network scoring of metabolic reaction activity on Recon3D: detecting label leakage, memorized…
Context-specific genome-scale metabolic modeling begins with scoring which of the approximately 10,600 human reactions are active in a patient's tumor. Methods in this literature are routinely benchmarked against activity labels obtained by thresholding the…
Boosting ALDOC enzyme improves heart function and reduces scar after heart attack in mice
On September 8, the journal Cell Death & Disease reported that researchers administered the Aldoc gene to adult male mice using an AAV9 vector immediately after inducing a myocardial infarction by ligating the left anterior descending coronary artery. Each group contained six mice, with controls receiving the same vector lacking Aldoc.
Twenty‑eight days later, the Aldoc‑treated hearts showed an ejection fraction of 56.74%, compared with 36.50% in controls.
The scar occupied 21.80% of the left‑ventricular area versus 41.95% in the control group.
The authors traced the effect to ALDOC’s interaction with the RNA‑processing protein RBM39; excess ALDOC reduced ubiquitination of RBM39, stabilizing it and thereby enhancing PI3K/AKT signaling, a pathway linked to cell division. In cell experiments, inhibiting PI3K/AKT weakened cardiomyocyte proliferation markers, and knocking down Rbm39 in the infarct model blunted the functional and scar‑size improvements conferred by ALDOC overexpression.
Earlier work showed that ALDOC levels are high in newborn mouse hearts, drop sharply by day seven, and rise again after cardiac injury; overexpressing ALDOC in neonatal cardiomyocytes increased glucose breakdown and proliferation signs, while its suppression impaired these processes. After apical resection in newborn mice, ALDOC knockdown led to poorer pumping function and larger scars at 28 days, linking the neonatal regenerative phenotype to the adult infarct rescue.
🔗 Read original →
On September 8, the journal Cell Death & Disease reported that researchers administered the Aldoc gene to adult male mice using an AAV9 vector immediately after inducing a myocardial infarction by ligating the left anterior descending coronary artery. Each group contained six mice, with controls receiving the same vector lacking Aldoc.
Twenty‑eight days later, the Aldoc‑treated hearts showed an ejection fraction of 56.74%, compared with 36.50% in controls.
The scar occupied 21.80% of the left‑ventricular area versus 41.95% in the control group.
The authors traced the effect to ALDOC’s interaction with the RNA‑processing protein RBM39; excess ALDOC reduced ubiquitination of RBM39, stabilizing it and thereby enhancing PI3K/AKT signaling, a pathway linked to cell division. In cell experiments, inhibiting PI3K/AKT weakened cardiomyocyte proliferation markers, and knocking down Rbm39 in the infarct model blunted the functional and scar‑size improvements conferred by ALDOC overexpression.
Earlier work showed that ALDOC levels are high in newborn mouse hearts, drop sharply by day seven, and rise again after cardiac injury; overexpressing ALDOC in neonatal cardiomyocytes increased glucose breakdown and proliferation signs, while its suppression impaired these processes. After apical resection in newborn mice, ALDOC knockdown led to poorer pumping function and larger scars at 28 days, linking the neonatal regenerative phenotype to the adult infarct rescue.
🔗 Read original →
Nature
ALDOC promotes cardiomyocyte proliferation and heart regeneration via suppressing RBM39-ubiquitination
Cell Death & Disease - ALDOC promotes cardiomyocyte proliferation and heart regeneration via suppressing RBM39-ubiquitination
Language models produce many erroneous references in cardiology reviews
The preprint examined 270 medical reviews, finding that 76.8% of erroneous links pointed to a real but different article; it was posted on medRxiv on 4 September. Researchers used three language models with web‑search to write cardiology reviews and checked 8050 references against PubMed.
One review placed PMID 17565084 next to a paper on cardiac resynchronization therapy, but that PubMed record actually opens a JAMA article about NIH grants for early‑career physician‑researchers. Verifying a reference requires first matching the identifier to the named article, then checking whether that article supports the claim it was cited for.
Each of the three models wrote ten reviews for each of nine cardiology topics; each review was 600–900 words long and contained 30 references. Of the 8036 references that had a PMID or DOI, problematic rates were 11.5% for GPT‑5.5, 29.2% for Claude Opus 4.8, and 29.6% for Gemini 3.5 Flash. Most problems were substitutions: a valid PMID or DOI attached to the wrong article, accounting for 76.8% of all errors.
The authors applied CoVe – a chain‑of‑verification method where the model breaks a reference into separate questions, retrieves the PubMed record by the given number, independently searches for the article by title and first author, then compares authors, journal, year, volume and pages. In a set of 270 references annotated by a cardiologist, CoVe found 60 of 62 problems, including all 51 substitutions and all four fabricated articles; three correct links were mistakenly flagged as problematic.
Separately, the team matched nine statements from the reviews with the abstracts of the cited articles. Reference checking consists of two steps: the identifier and title must lead to the same article, and that article must actually support the statement in the text.
🔗 Read original →
The preprint examined 270 medical reviews, finding that 76.8% of erroneous links pointed to a real but different article; it was posted on medRxiv on 4 September. Researchers used three language models with web‑search to write cardiology reviews and checked 8050 references against PubMed.
One review placed PMID 17565084 next to a paper on cardiac resynchronization therapy, but that PubMed record actually opens a JAMA article about NIH grants for early‑career physician‑researchers. Verifying a reference requires first matching the identifier to the named article, then checking whether that article supports the claim it was cited for.
Each of the three models wrote ten reviews for each of nine cardiology topics; each review was 600–900 words long and contained 30 references. Of the 8036 references that had a PMID or DOI, problematic rates were 11.5% for GPT‑5.5, 29.2% for Claude Opus 4.8, and 29.6% for Gemini 3.5 Flash. Most problems were substitutions: a valid PMID or DOI attached to the wrong article, accounting for 76.8% of all errors.
The authors applied CoVe – a chain‑of‑verification method where the model breaks a reference into separate questions, retrieves the PubMed record by the given number, independently searches for the article by title and first author, then compares authors, journal, year, volume and pages. In a set of 270 references annotated by a cardiologist, CoVe found 60 of 62 problems, including all 51 substitutions and all four fabricated articles; three correct links were mistakenly flagged as problematic.
Separately, the team matched nine statements from the reviews with the abstracts of the cited articles. Reference checking consists of two steps: the identifier and title must lead to the same article, and that article must actually support the statement in the text.
🔗 Read original →
medRxiv
Citation reliability of frontier large language models in medical writing and its automated verification
Large language models (LLMs) are increasingly used to draft medical manuscripts, yet their citations are unreliable and clinicians lack a validated way to verify them. We evaluated three frontier LLMs, Claude Opus 4.8, GPT-5.5, and Gemini 3.5 Flash, generating…
Navitoclax reduces scar size in aged mice after heart attack
In the preprint, Navitoclax reduced scar formation in 15‑month‑old mice following myocardial infarction. The authors tested the drug in middle‑aged male mice after ischemia‑reperfusion: the coronary artery was occluded for one hour, then reperfusion was restored.
Parallel to the mouse work, they re‑analyzed spatial
🔗 Read original →
In the preprint, Navitoclax reduced scar formation in 15‑month‑old mice following myocardial infarction. The authors tested the drug in middle‑aged male mice after ischemia‑reperfusion: the coronary artery was occluded for one hour, then reperfusion was restored.
Parallel to the mouse work, they re‑analyzed spatial
🔗 Read original →
Researchsquare
An error has occurred
Research Square is a preprint platform that makes research communication faster, fairer, and more useful.
AI‑Designed Drug Rentocertib Shows Anti‑Aging Effects in IPF Trial
The drug rentocertib, created by the AI‑driven company Insilico, was originally designed to treat idiopathic pulmonary fibrosis (IPF), a progressive lung‑scarring condition that usually leads to death within a few years. In addition to improving lung volume, the compound was found to lower systemic markers of biological aging across the whole body.
The first AI system mined medical records, blood analyses and scientific publications to pinpoint the
🔗 Read original →
The drug rentocertib, created by the AI‑driven company Insilico, was originally designed to treat idiopathic pulmonary fibrosis (IPF), a progressive lung‑scarring condition that usually leads to death within a few years. In addition to improving lung volume, the compound was found to lower systemic markers of biological aging across the whole body.
The first AI system mined medical records, blood analyses and scientific publications to pinpoint the
🔗 Read original →
Nature
Integration of proteomic aging clocks in a phase 2a clinical trial supports simultaneous geroprotective assessment
Nature Biotechnology - Six proteomic clocks are applied in a clinical trial to assess anti-aging effects.
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: dates: 8 September, 13 March, 1 September, maybe also mention of Millennium prize? Not a number. Also maybe mention of "OpenAI", "Levent Alpoge", "Tristan Bakmaster", "Sebastien Bubeck". Those are names; not numbers. Key numbers: maybe "8 September", "13 March", "1 September". Also maybe "thousand" (prize of millennium? Actually "премия тысячелетия" is Millennium Prize, $1 million? Not given). So we can bold dates.
We need at most 4-5 double asterisks. We'll choose: 8 September, 13 March, 1 September, maybe also Navier–Stokes? That's not a number. Could also bold Euler equation? Not needed. We'll keep three dates.
We need to wrap study/journal citations in single underscores. There's no explicit journal citation, but there is mention of "политике данных , обновлённой 13 марта" which is a policy update date, not a citation. There's also "OpenAI заявила , что её исследователи и программы-агенты не видели работ математиков ... до публикации". No journal. There's "В объявлении о решении Навье—Стокса OpenAI рассказала, что 1 сентября поручила агентам искать ответы на открытые задачи". Not a citation. There's no explicit study/journal. So maybe we don't need any underscores. But we could still wrap "политике данных" as not a citation. The rule: wrap study/journal citations and publication references in single underscores. If none, we can skip.
We need to ensure we don't wrap whole sentences in double asterisks. Only key facts.
We need to split into short paragraphs 2-3 sentences each.
Let's craft translation.
First, headline: something like "OpenAI says its models may have been improved by anonymized data from two mathematicians". Under 90 chars.
Check length: "OpenAI says its models may have been improved by anonymized data from two mathematicians" count: let's count roughly: OpenAI(5)+space+ says(4)=9, space+its(3)=13, space+models(6)=20, space+may(3)=24, space+have(4)=29, space+been(5)=35, space+improved(9)=45, space+by(2)=48, space+anonymized(11)=60, space+data(4)=65, space+from(4)=70, space+two(3)=74, space+mathematicians(13)=88. Good under 90.
Now blank line.
Now body paragraphs.…
🔗 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: dates: 8 September, 13 March, 1 September, maybe also mention of Millennium prize? Not a number. Also maybe mention of "OpenAI", "Levent Alpoge", "Tristan Bakmaster", "Sebastien Bubeck". Those are names; not numbers. Key numbers: maybe "8 September", "13 March", "1 September". Also maybe "thousand" (prize of millennium? Actually "премия тысячелетия" is Millennium Prize, $1 million? Not given). So we can bold dates.
We need at most 4-5 double asterisks. We'll choose: 8 September, 13 March, 1 September, maybe also Navier–Stokes? That's not a number. Could also bold Euler equation? Not needed. We'll keep three dates.
We need to wrap study/journal citations in single underscores. There's no explicit journal citation, but there is mention of "политике данных , обновлённой 13 марта" which is a policy update date, not a citation. There's also "OpenAI заявила , что её исследователи и программы-агенты не видели работ математиков ... до публикации". No journal. There's "В объявлении о решении Навье—Стокса OpenAI рассказала, что 1 сентября поручила агентам искать ответы на открытые задачи". Not a citation. There's no explicit study/journal. So maybe we don't need any underscores. But we could still wrap "политике данных" as not a citation. The rule: wrap study/journal citations and publication references in single underscores. If none, we can skip.
We need to ensure we don't wrap whole sentences in double asterisks. Only key facts.
We need to split into short paragraphs 2-3 sentences each.
Let's craft translation.
First, headline: something like "OpenAI says its models may have been improved by anonymized data from two mathematicians". Under 90 chars.
Check length: "OpenAI says its models may have been improved by anonymized data from two mathematicians" count: let's count roughly: OpenAI(5)+space+ says(4)=9, space+its(3)=13, space+models(6)=20, space+may(3)=24, space+have(4)=29, space+been(5)=35, space+improved(9)=45, space+by(2)=48, space+anonymized(11)=60, space+data(4)=65, space+from(4)=70, space+two(3)=74, space+mathematicians(13)=88. Good under 90.
Now blank line.
Now body paragraphs.…
🔗 Read original →
X (formerly Twitter)
OpenAI (@OpenAI) on X
We congratulate Levent Alpöge and Tristan Buckmaster on their remarkable mathematical work.
We (the researchers and the agents) did not see any of their work through any means until they released…
We (the researchers and the agents) did not see any of their work through any means until they released…
Higher infant cash payments linked to slower biological aging at age four
On September 8, Nature Human Behaviour published results from the US Baby’s First Years trial. Families with newborns were randomly assigned to receive either $333 or $20 per month. After four years, 735 children provided saliva samples for analysis.
Researchers measured DNA methylation and calculated the DunedinPACE score, which estimates the pace of aging from epigenetic marks. In the high‑payment group, DunedinPACE was lower by 0.17 standard deviation compared to the low‑payment group. No other epigenetic measures differed between the groups.
The trial enrolled a thousand mothers living below the US poverty line in 2018–2019; 400 received $333 and 600 received $20. Randomization ensured the only systematic difference was the assigned payment amount. By isolating income, the study shows that the early‑life environment can alter a DNA‑based aging biomarker in four‑year‑olds.
🔗 Read original →
On September 8, Nature Human Behaviour published results from the US Baby’s First Years trial. Families with newborns were randomly assigned to receive either $333 or $20 per month. After four years, 735 children provided saliva samples for analysis.
Researchers measured DNA methylation and calculated the DunedinPACE score, which estimates the pace of aging from epigenetic marks. In the high‑payment group, DunedinPACE was lower by 0.17 standard deviation compared to the low‑payment group. No other epigenetic measures differed between the groups.
The trial enrolled a thousand mothers living below the US poverty line in 2018–2019; 400 received $333 and 600 received $20. Randomization ensured the only systematic difference was the assigned payment amount. By isolating income, the study shows that the early‑life environment can alter a DNA‑based aging biomarker in four‑year‑olds.
🔗 Read original →
Nature
Effects of a randomized controlled trial of unconditional cash transfers on epigenetic measures of ageing and cognition in children…
Nature Human Behaviour - Raffington et al. find that providing low-income mothers with cash gifts for the first four years of their children’s lives appears to change the children’s...
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. No added commentary.
We need to identify important facts: dates: 8 September (OpenAI published article), 166 pages, Lean formalization, 10,000 agents, 88 hours search, 17 hours verification, 1934 Jean Leray, 2000 Clay Mathematics Institute, etc. Also maybe "positive viscosity", "smooth external force", etc. But we need to be selective: at most 4-5 double asterisks.
We need to decide which facts are most important: maybe the date of publication, the 166-page length, the 10k agents, 88 hours, 17 hours verification, the 1934 Leray, 2000 Clay Institute. That's many. We need only 4-5.
We can choose: 8 September, 166 pages, ≈10,000 agents, 88 hours, 17 hours maybe that's 5. Or we could include the Clay Institute year 2000 as important. But we need to limit.
We need to wrap each fact individually with double asterisks, not whole sentences. So we can embed them within sentences.
We also need to wrap study/journal citations in single underscores. There's mention of a preprint by Levanta Alpöge and Tristan Buckmaster? Actually "В препринте Левента Альпёге и Тристана Бакмастера сингулярность строится для уравнения Эйлера, где вязкости нет." That's a citation but not a journal. Could treat as preprint by Levanta Alpöge and Tristan Buckmaster maybe. Also maybe we need to cite the article itself? Not needed.
We need to preserve the content exactly, just translate.
Let's draft translation.
Headline: something like "OpenAI claims proof of Navier–Stokes singularity with viscous fluid". Must be under 90 chars.
Let's count: "OpenAI claims proof of Navier–Stokes singularity with viscous fluid" length ~ 55. Good.
Now blank line.
Now body paragraphs.
Paragraph 1: OpenAI published article on September 8 with claimed proof of singularity in Navier–Stokes equations, released full 166-page text and Lean formalization. The Lean language automatically checks reasoning steps.
We need to embed double asterisks for key numbers: maybe September 8, 166 pages. Also maybe Lean is not a number, but it's a name; we might not need to asterisk. We'll keep numbers.
Paragraph 2: According to the claimed theorem, for any positive viscosity the authors prescribe a smooth external force acting in a limited region of space and time, yielding a smooth three‑dimensional flow with finite kinetic energy that becomes unboundedly fast in finite time. The flow starts from rest, forms a vortex around an axis where fluid spirals inward and moves along the axis; the vortex core radius shrinks faster than its height, so speed grows while the volume shrinks enough to keep energy finite.
We might include positive viscosity maybe not needed. Could include finite kinetic energy but that's descriptive.
Paragraph 3: To avoid a non‑smooth remainder at the core boundary, the authors add short‑wavelength velocity oscillations around the core; the background vortex shear amplifies these waves, and their average momentum transport cancels the remainder. Subsequent corrections and spatial localisation keep the external force smooth even as the maximum speed grows without bound.…
🔗 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. No added commentary.
We need to identify important facts: dates: 8 September (OpenAI published article), 166 pages, Lean formalization, 10,000 agents, 88 hours search, 17 hours verification, 1934 Jean Leray, 2000 Clay Mathematics Institute, etc. Also maybe "positive viscosity", "smooth external force", etc. But we need to be selective: at most 4-5 double asterisks.
We need to decide which facts are most important: maybe the date of publication, the 166-page length, the 10k agents, 88 hours, 17 hours verification, the 1934 Leray, 2000 Clay Institute. That's many. We need only 4-5.
We can choose: 8 September, 166 pages, ≈10,000 agents, 88 hours, 17 hours maybe that's 5. Or we could include the Clay Institute year 2000 as important. But we need to limit.
We need to wrap each fact individually with double asterisks, not whole sentences. So we can embed them within sentences.
We also need to wrap study/journal citations in single underscores. There's mention of a preprint by Levanta Alpöge and Tristan Buckmaster? Actually "В препринте Левента Альпёге и Тристана Бакмастера сингулярность строится для уравнения Эйлера, где вязкости нет." That's a citation but not a journal. Could treat as preprint by Levanta Alpöge and Tristan Buckmaster maybe. Also maybe we need to cite the article itself? Not needed.
We need to preserve the content exactly, just translate.
Let's draft translation.
Headline: something like "OpenAI claims proof of Navier–Stokes singularity with viscous fluid". Must be under 90 chars.
Let's count: "OpenAI claims proof of Navier–Stokes singularity with viscous fluid" length ~ 55. Good.
Now blank line.
Now body paragraphs.
Paragraph 1: OpenAI published article on September 8 with claimed proof of singularity in Navier–Stokes equations, released full 166-page text and Lean formalization. The Lean language automatically checks reasoning steps.
We need to embed double asterisks for key numbers: maybe September 8, 166 pages. Also maybe Lean is not a number, but it's a name; we might not need to asterisk. We'll keep numbers.
Paragraph 2: According to the claimed theorem, for any positive viscosity the authors prescribe a smooth external force acting in a limited region of space and time, yielding a smooth three‑dimensional flow with finite kinetic energy that becomes unboundedly fast in finite time. The flow starts from rest, forms a vortex around an axis where fluid spirals inward and moves along the axis; the vortex core radius shrinks faster than its height, so speed grows while the volume shrinks enough to keep energy finite.
We might include positive viscosity maybe not needed. Could include finite kinetic energy but that's descriptive.
Paragraph 3: To avoid a non‑smooth remainder at the core boundary, the authors add short‑wavelength velocity oscillations around the core; the background vortex shear amplifies these waves, and their average momentum transport cancels the remainder. Subsequent corrections and spatial localisation keep the external force smooth even as the maximum speed grows without bound.…
🔗 Read original →
OpenAI
On the Navier–Stokes Millennium Prize Problem
We’re sharing an AI-generated solution to the Navier–Stokes Millennium Prize Problem, including a writeup and a formal proof in Lean.
Ovarian stiffness increases with age, linked to lower oocyte yield
A preprint published Preprint, September 8 measured ovarian tissue stiffness in 132 participants from fertility clinics. Researchers used shear‑wave elastography, an ultrasound technique that tracks tissue waves to gauge resistance to deformation. They compared the readings with ultrasound signs of tissue remodeling and with outcomes of IVF cycles.
In the first cohort, 16 women aged 33 or younger had an average stiffness of 13.3 kPa, while 16 women aged 37 or older averaged 18.8 kPa. With age, both the maximum value and the variability of the measurements increased. Mouse experiments from 2020 had suggested that old ovaries are about 2.5‑fold stiffer than young ones, and that collagenase can bring the tissue back toward youthful levels.
The age‑related rise in stiffness remained significant after adjusting for anti‑Müllerian hormone (AMH), a marker of follicular reserve, and for body‑mass index. A second cohort of 100 women aged 24‑45 showed the same three stiffness metrics climbing with age. These findings suggest that the ultrasonic measure reflects intrinsic tissue changes rather than hormonal or metabolic confounders.
To validate the ultrasound signal, researchers examined follicular fluid: the older group had lower levels of C6M, a fragment generated when collagen VI breaks down
🔗 Read original →
A preprint published Preprint, September 8 measured ovarian tissue stiffness in 132 participants from fertility clinics. Researchers used shear‑wave elastography, an ultrasound technique that tracks tissue waves to gauge resistance to deformation. They compared the readings with ultrasound signs of tissue remodeling and with outcomes of IVF cycles.
In the first cohort, 16 women aged 33 or younger had an average stiffness of 13.3 kPa, while 16 women aged 37 or older averaged 18.8 kPa. With age, both the maximum value and the variability of the measurements increased. Mouse experiments from 2020 had suggested that old ovaries are about 2.5‑fold stiffer than young ones, and that collagenase can bring the tissue back toward youthful levels.
The age‑related rise in stiffness remained significant after adjusting for anti‑Müllerian hormone (AMH), a marker of follicular reserve, and for body‑mass index. A second cohort of 100 women aged 24‑45 showed the same three stiffness metrics climbing with age. These findings suggest that the ultrasonic measure reflects intrinsic tissue changes rather than hormonal or metabolic confounders.
To validate the ultrasound signal, researchers examined follicular fluid: the older group had lower levels of C6M, a fragment generated when collagen VI breaks down
🔗 Read original →
bioRxiv
Multimodal profiling establishes ovarian fibrosis as a measurable and targetable hallmark of human reproductive aging
ABSTRACT Ovarian aging underpins infertility, systemic morbidity, and mortality in women. Stromal fibrosis is implicated in ovarian aging but has not been defined in the human ovary in situ. We integrated shear wave elastography (SWE), extracellular matrix…
Zebrafish heart repair uses distinct cardiomyocyte layers for growth and regeneration
On September 8, biologists described in eLife a thin layer of cardiomyocytes in zebrafish hearts that helps assemble the muscle wall and coronary vessels during growth.
After injury, other cardiomyocytes rebuild lost tissue; zebrafish hearts can regenerate muscle after ventricular tip removal; surviving cardiomyocytes change state, divide, and replace the lost tissue; the heart consists of several such layers.
Researchers mapped 1,668 cardiomyocytes of adult heart and found a group with active phlda2 gene; these cells lie in a thin layer between outer compact muscle and inner trabeculae.
They created a fish line where phlda2-active cells produce bacterial NTR enzyme and become sensitive to metronidazole; treatment removed about 96.9% of these cells from the ventricle.
In young fish, trabecular muscle area became roughly 54.7% smaller, outer compact layer lost proper organization, and coronary vessels fragmented; after 90 days the vascular pattern remained disrupted.
This layer participates in how the growing heart assembles muscle and vascular network.
In adult fish, ventricular tip was removed six days after treatment start; one week later the wound contained as many gata4-active cells (a marker of dividing cardiomyocytes during repair) as in controls.
By day 30 new muscle covered the injury; vessel density and scar area matched controls; the phlda2‑active layer did not recover even after 60 days, and its cells did not overlap with those activating gata4 during regeneration, indicating that heart growth and adult tissue repair rely on different cardiomyocyte populations.
🔗 Read original →
On September 8, biologists described in eLife a thin layer of cardiomyocytes in zebrafish hearts that helps assemble the muscle wall and coronary vessels during growth.
After injury, other cardiomyocytes rebuild lost tissue; zebrafish hearts can regenerate muscle after ventricular tip removal; surviving cardiomyocytes change state, divide, and replace the lost tissue; the heart consists of several such layers.
Researchers mapped 1,668 cardiomyocytes of adult heart and found a group with active phlda2 gene; these cells lie in a thin layer between outer compact muscle and inner trabeculae.
They created a fish line where phlda2-active cells produce bacterial NTR enzyme and become sensitive to metronidazole; treatment removed about 96.9% of these cells from the ventricle.
In young fish, trabecular muscle area became roughly 54.7% smaller, outer compact layer lost proper organization, and coronary vessels fragmented; after 90 days the vascular pattern remained disrupted.
This layer participates in how the growing heart assembles muscle and vascular network.
In adult fish, ventricular tip was removed six days after treatment start; one week later the wound contained as many gata4-active cells (a marker of dividing cardiomyocytes during repair) as in controls.
By day 30 new muscle covered the injury; vessel density and scar area matched controls; the phlda2‑active layer did not recover even after 60 days, and its cells did not overlap with those activating gata4 during regeneration, indicating that heart growth and adult tissue repair rely on different cardiomyocyte populations.
🔗 Read original →
Nature
Tbx5a lineage tracing shows cardiomyocyte plasticity during zebrafish heart regeneration
Nature Communications - It is not clear if it is the embryonic origin or anatomical location of cardiomyocytes that restrict their contribution to zebrafish heart regeneration. Here, the authors...
BioPharma APAC Maps Four Longevity Funding Channels in Asia‑Pacific
On September 9, BioPharma APAC released a map of four channels of longevity financing in the Asia‑Pacific region. The map separates funds that pay for therapy development, diagnostics, clinic services, and scientific projects linked to the region but financed or incorporated outside it. Venture‑tracker records show when an investor receives equity after a deal.
Payments from pharmaceutical companies for joint development, research grants, and clinic revenue from examinations go to different recipients and support distinct work, which the author keeps separate. Therapeutic programs in the map draw venture money, public capital, grants, and pharma partnership payments. Biological‑age diagnostics — tests that assess age‑related changes in the body — form a separate line.
Two additional lines cover clinics and services funded
🔗 Read original →
On September 9, BioPharma APAC released a map of four channels of longevity financing in the Asia‑Pacific region. The map separates funds that pay for therapy development, diagnostics, clinic services, and scientific projects linked to the region but financed or incorporated outside it. Venture‑tracker records show when an investor receives equity after a deal.
Payments from pharmaceutical companies for joint development, research grants, and clinic revenue from examinations go to different recipients and support distinct work, which the author keeps separate. Therapeutic programs in the map draw venture money, public capital, grants, and pharma partnership payments. Biological‑age diagnostics — tests that assess age‑related changes in the body — form a separate line.
Two additional lines cover clinics and services funded
🔗 Read original →
Nature
Recommendations for biomarker data collection in clinical trials by longevity biotechnology companies
npj Aging - Recommendations for biomarker data collection in clinical trials by longevity biotechnology companies
Radical Numerics launches Omnii for personalized mRNA cancer vaccine target selection
On September 8, Radical Numerics unveiled Omnii, a post‑trained model designed to pick targets for personalized mRNA cancer vaccines. The system takes a tumor’s mutational profile and the patient’s MHC variants, then selects vaccine targets and builds an mRNA cassette sequence.
Omnii first predicts whether a mutant peptide will be presented by the patient’s MHC class I molecules on the tumor cell surface. It then estimates the peptide’s immunogenicity, i.e., whether a T‑cell will recognize the presented fragment and mount a response. Only peptides that pass both presentation and immunogenicity checks can be included in the vaccine cassette.
In head‑to‑head comparisons, Omnii achieved an AUROC 0.939 for MHC class I presentation, outperforming BigMHC‑EL’s 0.933. For immunogenicity prediction, the model scored AUROC 0.750, compared with 0.558 for BigMHC‑IM.
In a simulation where 100 candidate peptides were evaluated, only 6% were truly immunogenic and the cassette had five slots. Omnii selected an average of 1.3 immunogenic targets per cassette, while BigMHC‑IM managed just 0.7.
After target selection, Omnii outputs the mRNA sequence encoding the chosen neoantigens. The model is currently available in research preview, and Radical Numerics is seeking partners in cancer immunology and vaccine development.
🔗 Read original →
On September 8, Radical Numerics unveiled Omnii, a post‑trained model designed to pick targets for personalized mRNA cancer vaccines. The system takes a tumor’s mutational profile and the patient’s MHC variants, then selects vaccine targets and builds an mRNA cassette sequence.
Omnii first predicts whether a mutant peptide will be presented by the patient’s MHC class I molecules on the tumor cell surface. It then estimates the peptide’s immunogenicity, i.e., whether a T‑cell will recognize the presented fragment and mount a response. Only peptides that pass both presentation and immunogenicity checks can be included in the vaccine cassette.
In head‑to‑head comparisons, Omnii achieved an AUROC 0.939 for MHC class I presentation, outperforming BigMHC‑EL’s 0.933. For immunogenicity prediction, the model scored AUROC 0.750, compared with 0.558 for BigMHC‑IM.
In a simulation where 100 candidate peptides were evaluated, only 6% were truly immunogenic and the cassette had five slots. Omnii selected an average of 1.3 immunogenic targets per cassette, while BigMHC‑IM managed just 0.7.
After target selection, Omnii outputs the mRNA sequence encoding the chosen neoantigens. The model is currently available in research preview, and Radical Numerics is seeking partners in cancer immunology and vaccine development.
🔗 Read original →
PubMed Central (PMC)
Artificial intelligence in peptide cancer vaccine design: from neoantigen discovery to immunogenicity prediction
Peptide-based cancer vaccines represent a promising immunotherapeutic strategy aimed at inducing tumor-specific immune responses through the targeting of tumor-associated antigens and neoantigens. Recent advances in next-generation sequencing and ...
IRIS RNA model predicts developmental signals to improve stem cell differentiation
The IRIS model uses RNA profiles to infer which developmental signals acted on a cell. It was described in Nature Methods, September 8 after being trained on human stem cells exposed to known combinations of six developmental signals and then tested on single‑cell data from mouse embryos.
An RNA profile records which genes are currently active, giving a snapshot of the cell’s state. A benchmark of 400 models on cell atlases showed that merely expanding collections of RNA profiles quickly ceases to improve predictive power.
IRIS first learns from cells with known signal combinations, then searches the RNA profile for the trace of a signaling pathway—the chain through which a cell responds to an external cue. Researchers cultured human embryonic stem cells, applied defined signal mixes at different stages, and measured single‑cell RNA after each condition to teach IRIS to recognize these traces.
When applied to mouse embryo data, IRIS recovered the expected order of signal combinations in lineages that give rise to the foregut and heart muscle. It then predicted heightened activity of WNT and BMP in the prospective respiratory mesenchyme surrounding the lung bud.
In mouse foregut culture, activating WNT expanded the domain where the respiratory mesenchymal markers Tbx4 and Foxf1 are expressed, while a WNT inhibitor erased Tbx4 expression. Guided by this finding, the authors added WNT stimulation on the fourth day of differentiation of human stem cells and maintained it, which raised TBX4 levels and kept intermediate WNT activity to preserve FOXF1.
🔗 Read original →
The IRIS model uses RNA profiles to infer which developmental signals acted on a cell. It was described in Nature Methods, September 8 after being trained on human stem cells exposed to known combinations of six developmental signals and then tested on single‑cell data from mouse embryos.
An RNA profile records which genes are currently active, giving a snapshot of the cell’s state. A benchmark of 400 models on cell atlases showed that merely expanding collections of RNA profiles quickly ceases to improve predictive power.
IRIS first learns from cells with known signal combinations, then searches the RNA profile for the trace of a signaling pathway—the chain through which a cell responds to an external cue. Researchers cultured human embryonic stem cells, applied defined signal mixes at different stages, and measured single‑cell RNA after each condition to teach IRIS to recognize these traces.
When applied to mouse embryo data, IRIS recovered the expected order of signal combinations in lineages that give rise to the foregut and heart muscle. It then predicted heightened activity of WNT and BMP in the prospective respiratory mesenchyme surrounding the lung bud.
In mouse foregut culture, activating WNT expanded the domain where the respiratory mesenchymal markers Tbx4 and Foxf1 are expressed, while a WNT inhibitor erased Tbx4 expression. Guided by this finding, the authors added WNT stimulation on the fourth day of differentiation of human stem cells and maintained it, which raised TBX4 levels and kept intermediate WNT activity to preserve FOXF1.
🔗 Read original →