Babak Alipanahi

Why All Mammograms Should Incorporate A.I. A very impressive body of evidence has accumulated

Is it time for a new standard of care when it comes to mammograms in the era of AI?
erictopol.substack.com/p/why-all-ma...

AI hallucinations in science manuscripts are a nuisance. Paranormal citations, or paracites, will be a nightmare.

www.biorxiv.org/content/10.6... (w/ @sina.bio & @lauraluebbert.com).

Time for a thread on our Christmas preprint “Origin and evolution of acrocentric chromosomes in human and great apes”. I had so much fun with this project and paper. It will be hard to summarize in a thread, but I’ll try www.biorxiv.org/content/10.6... [1/21]

A Closer Look at AUROC and AUPRC under Class Imbalance In machine learning (ML), a widespread claim is that the area under the precision-recall curve (AUPRC) is a superior metric for model comparison to the area under the receiver operating characteristic...

A good, enjoyable paper on AUROC vs AUPRC under class imbalance. In a nutshell, AUPRC's superiority is a myth.

AUROC with bootstrapping all the way!

arxiv.org/abs/2401.06091

TF-MINDI is out! A new method to learn cis-regulatory codes through rich embeddings of TF binding sites. TF-MINDI decomposes motif neighbourhoods, and works downstream of any sequence-to-function deep learning model. We deeply study the enhancer code in human neural development, check out the thread

Vaccines, the most impressive public health intervention in medical history, and where we could be headed if there was not efforts to negate truth, facts, and evidence
A great, open-access, review and perspective by @scientificdiscovery.dev

Now published in Algorithms for Molecular Biology: link.springer.com/article/10.1.... Key message: a tiny CNN model with 7k parameters can capture main splice signals across vertebrates+insect and halves the minimap2 & miniprot junction error rate. I always use this new feature now.

Now published in gigascience: academic.oup.com/gigascience/.... Key messages: SVs are highly enriched in low-complexity/tandem-repeat regions and are harder to call. They behave differently from transposon insertions. Always stratify if you study SVs.

Mapping isoforms and regulatory mechanisms from spatial transcriptomics data with SPLISOSM - Nature Biotechnology Differential isoform usage is identified with high statistical power from spatial transcriptomics data.

Excited to see this out www.nature.com/articles/s41...! Nonparametric kernel-based tests for spatially variable isoform usage in spatial transcriptomics. So many interesting examples in the CNS and cancer, we're only scratching the surface!

Uncovering the role of LINE-1 in the evolution of lung adenocarcinoma - Nature Lung adenocarcinomas bearing the ID2 mutational signature display increased LINE-1 retrotransposon activity, which contributes to their fast evolutionary dynamics and aggressive phenotype.

Nature research paper: Uncovering the role of LINE-1 in the evolution of lung adenocarcinoma

go.nature.com/4oUHIPb

That’s a wrap on #SABCS25! Thank you to Dr. Lee Schwartzberg for presenting data demonstrating our platform’s ability to detect early stage breast cancer with high accuracy. #AI #RNA #earlydetection

Learn more here: www.exai.bio/publications...

Risk-Based vs Annual Breast Cancer Screening This randomized clinical trial examines whether risk-based screening is a safe and effective alternative to annual mammography for detecting breast cancer in women 40 years and older.

Important new, large (N>28,000 women) randomized clinical trial of breast cancer screening: age-based vs risk-based by polygenic risk score, genomics
"opportunity to modernize screening"
jamanetwork.com/journals/jam...

Even though the highest-profile names in today’s corporate Cambridge are in biotech and software, the influx of defense startups hearkens back to an earlier era — which, in 1922, saw the birth of Raytheon, now synonymous with the old guard of defense contractors. www.thecrimson.com/article/2025...

SCIENCE SAVES LIVES.

Overall pediatric cancer survival rate increased from 63% in mid 1970s to 87%‼️between 2015 & 2021.

And this isn’t due to supplements, eating better or avoiding red food dye.

It’s due to science & industry working together to develop & approve therapies!

JASPAR 2026 is out 🎉

The new release massively expands the TF motif collections and adds a dedicated DeepLearning collection of motifs learned from deep learning models.

Database: jaspar.elixir.no
Paper (NAR): doi.org/10.1093/nar/...

🧵1/2

GitHub - lh3/human-asm: A collection of high-quality human genomes A collection of high-quality human genomes. Contribute to lh3/human-asm development by creating an account on GitHub.

579 high-quality human genomes from @humanpangenome.bsky.social, Arab Pangenome and individual papers (CHM13, CN1, KSA001, I002C, YAO and KOREF1). Sequences available in the AGC format (3.7GB) and FM-index in the ropebwt3 format (20.3GB). For details, see github.com/lh3/human-asm

This is also insightful:

"[...] we extracted the logic used by SpliceAI, Pangolin, and AlphaGenome to recognize exons within a fixed sequence context. Our analysis revealed fundamental limitations
in what models learn, including confounders and blind spots that compromise prediction reliability."

This figure explains the underlying concept pretty well:

A nice paper on distilling AI-based splicing models into much simpler additive models:

"[...] the distilled models achieve this without modeling RNA structure or feature interactions, indicating that [AI]-based splicing models recognize exons primarily through simple additive sequence features."

Abbott Nears Deal for Cancer Test Maker Exact Sciences Abbott Laboratories is nearing a potential acquisition of medical-testing company Exact Sciences Corp., in what would be its largest deal in nearly a decade, people familiar with the matter said.

Abbott Laboratories is nearing a potential acquisition of Exact Sciences Corp, in what would be its largest deal in nearly a decade, people familiar with the matter said. www.bloomberg.com/news/article...

Are you an early-stage graduate student (2nd or 3rd year) or early-stage postdoc based in the US or Canada, working primarily in Drosophila? Would you like to help improve the experience of all trainees working in Drosophila research? If so, read on.

(Please repost to reach a broad audience.)

Estimation and mapping of the missing heritability of human phenotypes - Nature WGS data were used from 347,630 individuals with European ancestry in the UK Biobank to obtain high-precision estimates of coding and non-coding rare variant heritability for 34 co...

First time on Bsky and first big announcement!

I am excited to announce that our new study explaining the missing heritability of many phenotypes using WGS data from ~347,000 UK Biobank participants has just been published in @Nature.

Our manuscript is here: www.nature.com/articles/s41....

A study found lead in popular protein powders. Here's why you shouldn't panic Consumer Reports expressed concern about high levels of lead in some two dozen protein powders, but only with repeated high exposure. Here's what to know before you make your next grocery run.

And then there is the whole "lead in protein powders" situation: www.npr.org/2025/10/16/n...

Yesterday our co-founder and CSO @babak-a.bsky.social presented at the Biotech-Pharma Statistics Workshop (BBSW) about the powerful combination of AI and cell-free RNA to detect early-stage lung cancer in the blood.

Caramelized Onion, Cranberry and Rosemary Tahchin  Recipe Tahchin is a Persian rice dish in which the rice is mixed with yogurt, oil, egg yolks and saffron and baked until a golden crust forms at the bottom (Persians refer to this as the tahdig) The rice on the inside becomes buttery and almost cake-like and is often layered with chicken and barberries, a tart dried fruit that has a beautiful crimson color This version incorporates common Thanksgiving ingredients like rosemary, sweet-tart cranberries and buttery onions to make a striking dish that feels more like a main than a side

This version of the Persian dish tahchin incorporates common Thanksgiving ingredients. It is deeply savory and buttery, like stuffing, and some may say even better because it has a whole lot more texture coming from the crispy rice that everyone will be fighting over.

Introducing Molview - the ipython/jupyter widget version of nano-protein-viewer🔍:

Neurodevelopmental Outcomes of 3-Year-Old Children Exposed to Maternal Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) Infection in Utero

I hate to say “I told you so…” but nevertheless: I told you so.

PRSformer: Disease Prediction from Million-Scale Individual Genotypes Predicting disease risk from DNA presents an unprecedented emerging challenge as biobanks approach population scale sizes (N>106 individuals) with ultra-high-dimensional features (L>105 genotypes). Cu...

Delighted to see our method, PRSformer, at #NeurIPS2025! PRSformer is AI model for population-scale disease-risk prediction from individual genomes. It lays the groundwork for phenome-wide risk prediction.

www.biorxiv.org/content/10.1...

Simultaneously comical and tragic.

Are you a PhD student interested in ML and biology or health? Come do an internship with me, @avapamini.bsky.social, Alex Lu, @lcrawford.bsky.social, or Kristen Severson at MSRNE!

Applications are due Dec 1: make sure you include a research statement!

jobs.careers.microsoft.com/global/en/jo...