I've written about race, genetic ancestry, analyses of large biobanks, and human history gusevlab.org/projects/hsq/#h… I'll summarize the key points here 🧵:

AmErIcAn DyNaMiSm

American century of embarrassment

It's not the universities that are failing young men.

Contra my previous post, it's kind of wild that NYC's Gifted & Talented program no treatment effect on standardized math test scores. 1/

This is still the best take on the ed policy debates:

IQ, giftedness, and the heritability thereof is maybe the best example of a divide between *rhetorical* empiricists/technocrats and *actual* experts. The lib pundits are still in thrall to The Bell Curve; actual geneticists & neurologists have utterly shredded it

😮

1. If you ask an LLM to pick the best normalization method for single-cell RNA-seq data, what does it suggest, and why? The answer points to an under-looked aspect of doing science in the age of AI.

Self-supervised learning on observational scRNAseq especially with naive reconstruction or masking losses will never deliver a causal model of gene regulation. It will learn statistical structure in expression space, not the mechanisms that generate regulatory change. 1/

Are colorectal cancer rates really on the rise in the young?? The shape of the incidence curves suggest not. Something happened around 2013. That something was the ICD classification system allowing neuroendocrine tumors of the appendix to be considered malignant colorectal cancers. [1/2]

Scarr-Rowe hypothesis in action

Two years ago the best virtual cell model was ridge regression and yesterday the best virtual cell model was ... ridge regression. But sure, throw a billion more parameters at the problem, no one is stopping you.

Arguably the most boring step in genomics is the first one: normalization. Settled science. Scale log. Move on. Except that here's been a huge blind spot in the field. And it matters for AIxBio. A 🧵about what I think may be one of the most important papers I've written. 1/

My first main use of Mythos-backed Fable model is to go through older AI-driven vibe-coded analysis that may have been sketch and figure out if they were actually right or not. So far very helpful! (1/2)

No scaling laws for single-cell foundation models: when bigger atlases stop teaching the model anything In language and vision, the recipe has been simple: more data, bigger models, better performance. Single-cell biology borrowed that playbook. Foundation models for transcriptomics jumped from 1 million cells to atlases of over 100 million, on the assumption that scale would unlock the same gains. Alan DenAdel and coauthors put that assumption to the test, and the result is sobering. Working from a 22.2-million-cell corpus, they pretrained 400 models across five architectures (from PCA and a variational autoencoder up to the Geneformer transformer) and ran 6,400 evaluation experiments. They varied not just dataset size (1% to 75%) but also diversity, using cell-type re-weighting and geometric sketching to deliberately enrich rare cell types and transcriptional states. The finding: performance saturates almost immediately. On cell-type classification, batch integration, and perturbation prediction, most models hit their ceiling at roughly 1% of the corpus, about 200,000 cells. Beyond that, adding millions more cells changed essentially nothing. More diversity didn't help. Even spiking in genome-scale Perturb-seq data, to give the models perturbed phenotypes rather than just healthy ones, failed to move the needle. Larger models did score better overall, but they too plateaued early on data. Two points stood out. Simple baselines (PCA, logistic regression) often matched or beat the transformers. And the strongest model, SCimilarity, won not because of size but because its contrastive training objective is aligned with the downstream task. For single-cell data, what you train on and how you frame the objective matters far more than how much you collect. This reframes a quiet but expensive habit. In drug discovery, biotech, and any pipeline leaning on cell atlases, the instinct to keep scaling pretraining corpora may be burning compute for no return. The real leverage sits elsewhere: curating high-quality, task-relevant data and matching the training objective to the actual question you're trying to answer. Paper: DenAdel et al., journal license | doi.org/10.1038/s41592-026-0…