We like producing and contributing to academic research. We also have a strong interest in writing software, tools, and platforms that help researchers be more productive - e.g. see kotsu, a lightweight framework written in Python for structuring ML model validation and comparison.
Here’s a collection of our peer-reviewed publications. If you’d like to discuss any of this work and especially if you want to explore research collaborations, drop us an email, we’d be stoked to hear from you!
MSA Pairing Transformer: protein interaction partner prediction with few-shot contrastive learning
ICML 2024 Workshop on Efficient and Accessible Foundation Models for Biological Discovery, June 2024, dvp authors: Alex, Dan
We tackle protein sequence pairing within interacting protein families by fine-tuning the MSA Transformer using contrastive learning applied to embeddings from scrambled single-chain multiple sequence alignments. Our approach achieves high pairing accuracy even within small sets of pairable sequences, outperforming prior co-evolutionary statistical methods. The paired cross-chain MSAs demonstrate stronger encoding of interface contacts compared to established heuristic approaches, suggesting potential for extending co-evolutionary analysis to protein-protein interactions.
Keywords: protein-folding, protein language model, protein-protein interaction, MSA, contrastive learning