Reproduce “Pairing interacting protein sequences using masked language modeling” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Umberto Lupo, Damiano Sgarbossa, Anne-Florence Bitbol
A differentiable framework, DiffPALM, infers interacting protein sequence pairings by adapting the Masked Language Modeling objective from MSA Transformer. This method demonstrates superior performance over traditional coevolutionary techniques on shallow multiple sequence alignments and enhances the accuracy of AlphaFold-Multimer's complex structure predictions for eukaryotic proteins.
Key insights
- MSA Transformer, despite being pre-trained solely on single-chain MSAs, implicitly captures inter-chain coevolutionary signals, allowing it to accurately predict inter-protein contacts and score potential pairings in a "zero-shot" manner.
- Formulating the paralog matching problem in a differentiable manner, combined with robust optimization techniques like Gumbel noise and multi-run aggregation, enables efficient navigation of the complex permutation landscape to find optimal pairings.
- The iterative incorporation of high-confidence predicted pairs as "positive examples" further refines the pairing process, significantly improving precision and demonstrating the method's ability to integrate prior knowledge.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.