COMICS: A pipeline for the composite identification of selection across multiple genomic scans using Invariant Coordinate Selection in R

Abstract

Identifying loci that are under selection versus those that are evolving neutrally is a common challenge in evolutionary genetics. Moreover, with the increase in sequence data, genomic studies have begun to incorporate the use of multiple methods to identify candidate loci under selection. Composite methods are usually implemented to transform the data into a multi-dimensional scatter where outliers are identified using a distance metric, the most common being Mahalanobis distance. However, studies have shown that the power of Mahalanobis distance reduces as the number of dimensions increases. Because the number of methods for detecting selection continue to grow, this is an undesirable feature of Mahalanobis distance. Other composite methods such as invariant coordinate selection (ICS) have proven to be a robust method for identifying outliers in multi-dimensional space; though, this method has not been implemented for genomic data. Here we use simulated genomic data to test the performance of ICS in identifying outlier loci from multiple selection scans and compare the results to the performance of Mahalanobis distances. We show that the ICS outperformed Mahalanobis distance in all aspects including false positives, false negatives, and recall. Furthermore, ICS also performed better when identifying loci with weaker selection coefficients. We also introduce a pipeline in a R-Shiny smart wrapper environment that implements the ICS on multiple scans of selection. Importantly, we show that the ICS is a robust method for identifying outliers in multi-dimensional space and recommend its use for studies aimed at identifying loci under selection in the genome.