Prots2Net: a PPIN predictor of a proteome or a metaproteome sample

Abstract

All molecular functions and biological processes are carried out by groups of proteins that interact to each other. Proteins interactions are modeled by simple networks called Protein-Protein Interaction Networks (PPINs) whose nodes are proteins and whose edges are the protein-protein interactions. PPINs are broadly accepted to model the protein's functional relations, and their analysis has become a key ingredient in the study of protein functions. New proteins are collected every day from metaproteomic data, and their functional relations must be obtained with high-throughput technology. Retrieving protein-protein interaction data experimentally is a very high time-consuming and labor-intensive task. Consequently, in the last years, the biological community is looking for computational methods to correctly predict PPIs. We present here Prots2Net, a tool designed to predict the PPIs of a proteome or a metaproteome sample. Our prediction model is a multilayer perceptron neural network that uses protein sequence information only from the input proteins and interaction information from the STRING database. To train the model, Prots2Net explores the PPIs retrieved from the STRING database of two selected species. The tests, reported here on the Yeast and the Human datasets, show that Prots2Net performs better than the previous prediction methods that used protein sequence information only. Therefore, considering the information of PPI data available on the STRING database improves the PPI prediction. Prots2Net is available at https://github.com/adriaalcala/prots2net