Combining structural modeling and deep learning to calculate the E. coli protein interactome and functional networks.

Abstract

We report on the integration of three methods that predict, on a proteome-wide scale, whether two proteins are likely to form a binary complex. The methods include PrePPI, which uses three-dimensional structure information as a basis for predictions, Topsy-Turvy, which uses a protein language model, and ZEPPI, which uses evolutionary information to evaluate protein-protein interfaces. Testing on the high-quality HINT database of binary PPIs reveals that the integrated method has better performance and identifies more high-confidence interactions than any of the component methods. The AF3Complex algorithm is used to predict the structures of 374 PPIs with a large fraction having at least partially overlapping interfaces with PrePPI models of the same complex. Clustering of the high-confidence E. coli interactome yields 385 subnetworks which have high functional coherence. Biological insights derived from the subnetworks, including the annotation of proteins of unknown function, are discussed in detail.