Seq-InSite: sequence supersedes structure for protein interaction site prediction

Abstract

Proteins accomplish cellular functions by interacting with each other, which makes the prediction of interaction sites a fundamental problem. Computational prediction of the interaction sites has been studied extensively, with the structure-based programs being the most accurate, while the sequence-based ones being much more widely applicable, as the sequences available outnumber the structures by two orders of magnitude. We provide here the first solution that achieves both goals. Our new sequence-based program, Seq-InSite, greatly surpasses the performance of sequence-based models, matching the quality of state-of-the-art structure-based predictors, thus effectively superseding the need for models requiring structure. Seq-InSite is illustrated using an analysis of four protein sequences. Seq-InSite is freely available as a web server atseq-insite.csd.uwo.caand as free source code, including trained models and all datasets used for training and testing, atgithub.com/lucian-ilie/seq-insite.

Scoring alignments by embedding vector similarity

“…Many protein embeddings have been proposed, including ProtVec [4], SeqVec [10], PRoBERTa [22], MSA-transformer [27], ESM-1b and ESM2 [28], and the models in the ProtTrans project: ProtTXL, ProtBert, ProtXLNet, ProtAlbert, ProtElectra, and ProtT5 [8]. These protein embeddings have many important applications to various areas, such as: structure prediction [29, 33, 15], function prediction [16, 9, 17], interaction site prediction [13, 35, 12], etc. In many applications they help achieving state-of-the-art performance.…”