Protein language models can learn homo-oligomer contacts from single sequences

anshulkundaje · x · 2026-07-22

A reposted discussion highlights a paper finding that protein language models trained only on individual sequences can implicitly learn interface contacts in homo-oligomeric assemblies.

As model scale increases, interface signals become more visible. The authors also note some PDB-labeled homo-oligomers do not show the expected signals, which led them to investigate when these contacts emerge—and when they are absent.

Related event: Protein Language Models Emerge Interface Signals from Single Sequences(2 posts)→

Original post →

More from Research

Research channel →