POTION: An End-to-End Pipeline for Positive Darwinian Selection Detection in Genomic Scale Data through Phylogenetic Comparison of Protein-Coding Genes

Detection of genes evolving under positive Darwinian evolution in genomic scale data is nowadays a prevailing strategy in comparative genomics studies to identify genes potentially involved in adaptation processes. In the specific case of parasitic organisms, such strategy could be used to detect genes responsible for parasitic phenotype, which are natural candidates for intervention. Despite the large number of studies aiming to detect and contextualize such gene sets, there is virtually no software available to perform this task in a general, automatic, large-scale and reliable manner. This certainly occurs due to the computational challenges involved in this task, such as the appropriate modeling of data under analysis, the computation time to perform several of the required steps when dealing with genomic scale data and the highly error-prone nature of the sequence and alignment data structures needed for genome-wide positive selection detection. We present POTION, an open source, modular and end-to-end software for genome-wide detection of positive Darwinian selection in groups of homologous coding sequences. Our software represents a key step towards genomic-scale, automated detection of positive selection, from predicted coding sequences and their homology relationships to high-quality groups of positively selected genes. POTION reduces false positives through several sophisticated sequence and group filters based on numeric, phylogenetic, quality and conservation criteria to remove spurious data and through multiple hypothesis corrections, and considerably reduces computation time due to a heavily parallelized design. Our software achieved a high classification performance when used to evaluate a curated dataset of homologous genes previously surveyed for positive selection. When used to analyze predicted homologous groups of 19 strains of Mycobacterium tuberculosis – an important pathogen of bovines, swines and other livestock - as a case study we demonstrated the filters implemented in POTION to remove noisy data that commonly inflate errors in positive selection detection. A thorough literature review found no other software similar to POTION in terms of customization, scale, reliability and automation. To the best of our knowledge, POTION is a first of its kind tool that allows users to construct and check hypotheses regarding the occurrence of site-based evidence of positive selection in genomic-scale data within a feasible time frame.