Protein Function Prediction from Sequence and Structure

What does a protein do - and can its sequence and its predicted 3-D structure tell us? This project predicts 30 Gene Ontology molecular functions (catalytic activity, DNA binding, transporter, receptor ...) for reviewed human proteins from the ESM-2 protein-language-model embedding and the AlphaFold DB structure, fused early, late and in a graph network on the residue contact map, and compares them with term frequency and Smith-Waterman homology transfer. Under a leakage-safe identity-cluster split early fusion reaches Fmax 0.565 (sequence only 0.558, structure only 0.484), while homology transfer falls from 0.461 on a random split to 0.325; on yeast proteins the models still reach 0.565 - 0.579.

Category: Bioinformatics - Proteomics and Protein Bioinformatics

Proteomics & Protein Bioinformatics Project B23 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.