David Bonet

My group's paper out on the cover of Science! Native Hawaiians are most closely related to the Rapanui of distant Easter Island 😮 Proud of it my students, the two co-first authors Cole Shanks & former MS student @edwardhuangc, and 3rd author @chrisoutho. science.org/eprint/4HAYG3KTP…

Fst is a key genetic differentiation metric, but it emphasizes common variants over rare. In @ieeeICASSP out now @geletavc introduces Fq, which can be tuned across the allele freq spectrum. Fst & Shannon differentiation fall out of it as special cases. arxiv.org/abs/2511.03063

ARGformer is out in Bioinformatics! doi.org/10.1093/bioinformati… Just as PCA produces a low-dimensional representation of population structure from genotypes, ARGformer can do this now for ARGs...a representation that comes now directly from inferred genealogical topologies & at a haplotype level across each individual's genome. We do this by turning local genealogical paths into tokens & learning dense embeddings, similar to language embeddings, to visualize, cluster & query. Code github.com/AI-sandbox/ARGfor… Team @d_bonet, Cole Shanks, @marcal_cc, @jordiatjhu

Congrats to first author @d_bonet, who presented ARGformer yesterday in a talk at the European Conference on Computational Biology @ECCBinfo in Geneva!

Gnomix is published! nature.com/articles/s41467-0… Still the most accurate & fastest local ancestry inference (LAI) method, bests Flare and others (Recomb-Mix & Orchestra) released since our preprint. Easy use github.com/AI-sandbox/gnomix Leads @HelgiHilmars @arvind0422 @BarrabesMiriam

@d_bonet and @marcal_cc representing the lab at the #KDD2026 conference in Korea @kdd_news, presenting their iLTM paper!

Just got our Genome Research article by Margarita Geleta @geletavc. 📬 I love hard copies...still here's a link genome.cshlp.org/content/36/… Rita's article was a tour de force of genomic uses for the autoencoder architecture: compression, dim reduction, ancestry, synthetic genomes.

Our paper, arxiv.org/abs/2511.15941 Our repo, github.com/AI-sandbox/iLTM Us, @d_bonet holding his award for our previous tabular model Hyperfast arxiv.org/abs/2402.14335 with @_danielmas and me & right adding co-author @marcal_cc while enjoying Hawaii for @PacSymBiocomp conference!

Some data isn't fit for LLMs, especially many biomedical datasets & anything using spreadsheets. Large Tabular Models (LTMs) are the way to go! Our new LTM (iLTM accepted #KDD2026) is now literally in the dictionary! Thanks @MerriamWebster! merriam-webster.com/dictiona…

Our ADAMIXTURE method is now out in Bioinformatics! Scaling genetic clustering (ADMIXTURE/STRUCTURE-style unsupervised clustering) is now a solved problem imo. ADAMIXTURE can cluster all of @uk_biobank (half a million genomes) in just over an hour with likelihoods as good as ADMIXTURE, which would take weeks. @JoanSaurina did an amazing job efficiently coding this! ADAMIXTURE is faster than any existing method and with none of their compromises. For comparison, the @AllofUsResearch paper clustering dropped to only 150k variants (far below what would be sufficient for fine scale local ancestry) then projected down to only 16 dimensions 😳 (and those were supervised PCs, so the method was no longer unsupervised, but confined to the 3k reference samples' ancestry space). Other methods make other compromises and are all still much slower. doi.org/10.1093/bioinformati…

Proud of @BarjonCar who just earned her PhD after getting the cover of Nature Medicine earlier this spring! In her study we show clinically relevant allele frequencies can vary regionally across Mexico by more than even between many global populations. nature.com/articles/s41591-0…

ARGformer is designed to do for ARG genealogies what PCA has long done for genotypes: produce a low-dimensional representation of population structure. But now this comes directly from inferred genealogical topologies & at a haplotype level across each individual genome.

@d_bonet is presenting ARGformer tonight at #PEQG26! We released ARGformer this winter bc ARGs are huge & hard to use. It turns local genealogical paths into tokens & learns dense embeddings, similar to language embeddings, to visualize, cluster & query. biorxiv.org/content/10.64898…