Researchers Debate Using LLMs to Crack the Genomic Motif Annotation Bottleneck

jatin_n0 · x · 2026-09-24

Computational biologist Anshul Kundaje and jatinn0 discussed using LLMs for functional annotation of genomic motifs. Kundaje clarified this isn't motif discovery but biologically annotating what a motif is, does, and binds to — a multimodal reasoning task well suited to LLMs with proper training that could unlock a massive bottleneck in genomic sequence annotation.

jatinn0 hoped that simply adding more sequence data during LLM training could yield functional roles of motifs, but Kundaje countered that annotation requires external information not in the sequence, such as protein availability in specific cell types. He teased that their ENCODE GRAMMAR motif compendium preprint will be out in a few weeks, along with a blog post detailing the challenge.

Original post →

Researchers Debate Using LLMs to Crack the Genomic Motif Annotation Bottleneck

More from AGI Musings