ProPythia: A Python package for protein classification based on machine and deep learning

The field of protein data mining has been growing rapidly in the last years. To characterize proteins and determine their function from their amino acid sequences are challenging and long-standing problems, where Bioinformatics and Machine Learning have an emergent role. A myriad of machine and deep learning algorithms have been applied in these tasks with exciting results. However, tools and platforms to calculate protein features and perform both Machine Learning (ML) and Deep Learning (DL) pipelines, taking as inputs protein sequences, are still lacking and have their limitations in terms of performance, user-friendliness and restricted domains of application. Here, to address these limitations, we propose ProPythia, a generic and modular Python package that allows to easily deploy ML and DL approaches for a plethora of problems in protein sequence analysis and classification. It facilitates the implementation, comparison and validation of the major tasks in ML or DL pipelines including modules to read and alter sequences, calculate protein features, preprocess datasets, execute feature selection and dimensionality reduction, perform clustering and manifold analysis, as well as to train and optimize ML/DL models and use them to make predictions. ProPythia has an adaptable modular architecture being a versatile and easy-to-use tool, which will be useful to transform protein data in valuable knowledge even for people not familiarized with ML code. This platform was tested in several applications comparing with results from literature. Here, we illustrate its applicability in two cases studies: the prediction of antimicrobial peptides and the prediction of enzymes Enzyme commission (EC) numbers. Furthermore, we assess the performance of the different descriptors on four different protein classification challenges. Its source code and documentation, including an user guide and case studies are freely available at https://github.com/BioSystemsUM/propythia.(c) 2021 Elsevier B.V. All rights reserved.

Automated summary

The field of protein data mining has been rapidly growing, but tools and platforms for protein sequence analysis and classification still have limitations. To address these limitations, ProPythia, a versatile and easy-to-use Python package, has been proposed to deploy machine learning and deep learning approaches for protein sequence analysis and classification.