MSA can still beat big models when protein families have very few sequences
sokrypton · x · 2026-07-22
MSA will still beat bigModel in the hardest cases, according to the thread:
- The key advantage appears when a protein family has very few sequences. In that regime, multiple sequence alignments can still provide signal that a pLM struggles to infer alone.
- For large protein families, the author argues you can often just pull related MSAs from AFDB and place the query sequence on top.
- The comparison is framed as smallMSA vs bigModel rather than a blanket “MSA vs model” debate.
- A table credited to @yoakiyama is cited as supporting evidence.
The broader point: alignment-based priors still matter a lot when data are scarce, even as large protein language models improve.
Related event: Protein Modeling: Small MSA vs Large Language Models(2 posts)→
More from Research
- Steve Hsu on OpenAI's Navier-Stokes paper: 'maybe beyond what a human could have come up with' — LucaAmb · 2026-10-27
- Skyfall AI's SKY-001 world model turns sketches into CAD builds in 3.9s, vs 199s for GPT-6 Astra — rohanpaul_ai · 2026-10-27
- Physicists Let AI Pick Its Own Problem—It Wrote a Publishable Paper in 2 Hours — burny_tech · 2026-10-27
- Researcher flags unresolved data transparency issues in compositionality evals — najoungkim · 2026-10-27
- Separate Forms from Estimators: Why Steer by Difference of Means When You Can Use SGD? — johnhewtt · 2026-10-27
- EMNLP paper probes why some LLM biases survive rewording while others don't — radamihalcea · 2026-10-27