Interrelational Proteomic Sequence Features Enhance Predictive Modeling: Application to COVID-19 Severity

Abstract

Background: Comparing biological properties among related proteins has traditionally benefited research in areas such as biomedicine, phylogeny and evolution. Moreover, these kinds of properties have significantly increased as a result of the development of open-access resources and databases. In this context, the multiple sequence alignment (MSA) methods continue to be extensively applied to compare protein sequences and to identify evolutionarily conserved regions. Methods: In this work, we present INPROF, a novel web server designed to centralize and automate the computation of a large collection of features derived from protein sequences. This tool allows us to deeply analyze protein relationships and their conserved information by comparing them through their MSA. Specifically, this platform computes up to 46 different metrics including information at several proteomic levels (categories) like sequences, structures, domains or ontological terms. INPROF was designed to enable bioinformaticians and researchers to create a complete catalogue of features for subsequent prediction and artificial intelligence modeling based on proteins. The INPROF web server and source code are freely available. Results: INPROF were validated by predicting disease’s severity in several RNA-Seq datasets from COVID-19 patients. Specifically, INPROF were extracted from canonical proteins related to differentially expressed genes. Classification performance proved that INPROF were able to accurately classify COVID-19 severity, even outperforming classical classification with expression data. Conclusions: INPROF web server is a flexible platform designed to compute 46 quantitative metrics describing protein interactions which provide biologically meaningful characteristics applicable to downstream classification and prediction algorithms.