bioRxiv2024
MotivationHigh-dimensional Bulk RNA sequencing (RNAseq) datasets pose a considerable challenge in identifying biologically relevant features for downstream analyses and data mining efforts. The standard approach involves differential gene expression (DGE) analysis, but its effectiveness can be limited depending on the data due to its univariate nature. In complex datasets, an alternative approach involves employing a variety of machine learning (ML) tools, which attempt to understand non-linear relationships between features and focus on generalizability rather than statistical significance. This approach will result in the generation of multiple feature lists, which might exhibit similarities in terms of classification performance metrics. Therefore, there is an urgent need for a cohesive workflow that seamlessly integrates robust feature selection using diverse ML methods while also evaluating the biological relevance of the resulting feature lists. This combined approach would enable the prioritization of the best-performing list, considering both sets of criteria. ResultsWe introduce GeneSelectR, an open-source R package that innovatively combines ML and bioinformatic data mining approaches for enhanced feature selection. With GeneSelectR, features can be selected from a normalized RNAseq dataset with a variety of ML methods and user-defined parameters. This is followed by an assessment of their biological relevance with Gene Ontology (GO) enrichment analysis, along with a semantic similarity analysis of the resulting GO terms. Additionally, similarity coefficients and fractions of the GO terms of interest are calculated. With this, GeneSelectR optimizes ML performance and rigorously assesses the biological relevance of the various lists, offering a means to prioritize feature lists with regard to the biological question. When applied to the TCGA-BRCA dataset, the GeneSelectR workflow generated several feature lists using different ML methods and a DGE analysis. By leveraging the various functions in GeneSelectR, the different lists could be evaluated based on both ML performance and biological relevance. This comprehensive evaluation facilitated the selection of the best-performing list, which exhibited both strong machine learning performance and high relevance to the biological question while maintaining a manageable number of highly specific features. AvailabilityThe package is available on CRAN. To install it, run: install.packages( GeneSelectR) Contactdzhakparov@gmail.com Supplementary informationSupplementary data are available at Bioinformatics online.