CIRPIN trains equivariant GNNs on circular permutations to fix protein structure search blind spots
bravo_abad · x · 2026-09-08
Kolodziej and coauthors introduce CIRPIN, addressing a blind spot in protein structure search: two proteins can share essentially the same 3D fold even when their chains start and end at different positions, yet search model embeddings still encode chain position and miss the relationship.
- They train an E(n)-equivariant graph neural network on synthetic circular permutations of protein structures
- Supervised contrastive learning makes different reorderings of the same structural topology land close together in embedding space
- Key insight: they don't discard positional information—they teach the model which changes in it shouldn't matter
This is enough to uncover protein relationships hidden from conventional structure search.
More from Research
- Anima Anandkumar: neural operators cracked real-world turbulence where LLM proofs fall short — AnimaAnandkumar · 2026-09-08
- Terry Tao breaks down new finite-time blowup advances toward Navier-Stokes regularity — Singularitarian · 2026-09-08
- Terence Tao blogs on Buckmaster and Alpöge's blow-up solutions to fluid equations — LucaAmb · 2026-09-08
- ECCV 2026 to Host Interactive Social Avatars Workshop With Keynotes From Andrew Zisserman and More — JonathonLuiten · 2026-09-08
- Building an active monitor for a black-box LLM: where to draw the automation line — Particular-Roof4257 · 2026-09-08
- NYU mathematician says LLMs helped crack fluid blowup problems in a month, accuses OpenAI of scoop and authorship pressure — dotey · 2026-09-08