Web-based tool for the annotation of pathological variants on proteins: PMut 2017 update

Abstract

Assessing the impact of amino acid mutations in human health is an important challenge in biomedical research. As sequencing technologies are more available, and more individual genomes become accessible, the number of identified variants has dramatically increased. PMut, released back in 2005 [1], has been one of the popular predictors in this field. PMut was a neural-network-based classifier using sequence data to provide a pathology score for point mutations in proteins. We now release a new, revised, and much more powerful version of PMut. It features PyMut prediction engine, a Python module that includes numerous machine learning capabilities aimed at the analysis of protein variant pathology annotation. We also release PMut2017 predictor, a full update of the PMut predictor based on the SwissVar [2] variation database. It achieves an accuracy of 82% and a Matthews Correlation Coefficient (MCC) of 0.62, and matches the most popular predictors’ performance. The engine is implemented in Python using MongoDB engine for data management. It has been adapted to run at the HPC level to cover large scale annotation projects.