DiffPaSS – Differentiable and scalable pairing of biological sequences using soft scores

Umberto Lupo, Damiano Sgarbossa, Martina Milighetti, Anne-Florence Bitbol

Track: Machine learning: computational method and/or computational results

Keywords: Protein-protein interactions, PPI, Paralog matching, Paired MSA, MSA, Multiple Sequence Alignment, Optimization, Graph alignment, Mutual Information, Proteins, TCR, AlphaFold, AlphaFold-Multimer, protein complexes

TL;DR: A family of fast, flexible, scalable and hyperparameter-free algorithms for pairing biological sequences.

Abstract:

Identifying interacting sequences from two sets of potential partners has important applications in computational biology.
Several methods have been developed to address this problem, applying different approximate optimization methods to different scores.
We introduce DiffPaSS, a framework for flexible, fast, scalable, and hyperparameter-free optimization for pairing interacting biological sequences, which can be applied to a wide variety of scores.
DiffPaSS consistently finds strong score optima, outperforming existing algorithms for optimizing the same scores.

Submission Number: 87