ARGformer lands in Bioinformatics: turning genealogical topologies into language-style embeddings

anshulkundaje · x · 2026-09-04

ARGformer, published in Bioinformatics, learns haplotype-level genome representations directly from inferred ancestral recombination graphs (ARGs), analogous to how PCA extracts population structure from genotypes.

It converts local genealogical paths into tokens and learns dense embeddings similar to language embeddings, enabling visualization, clustering and querying. Code is open-sourced.

Original post →

More from Research

Research channel →