BioFeatureFinder: Flexible, unbiased analysis of biological characteristics associated with genomic regions

Abstract BioFeatureFinder is a novel algorithm which allows analyses of many biological genomic landmarks (including alternatively spliced exons, DNA/RNA-binding protein binding sites, and gene/transcript functional elements, nucleotide content, conservation, k-mers, secondary structure) to identify distinguishing features. BFF uses a flexible underlying model that combines classical statistical tests with Big Data machine-learning strategies. The model is fed with thousands of biological characteristics (features) that are used to interpret category labels in genomic ranges or numerical scales from genome graphs. Our results show that it is a reliable platform for analyzing large-scale datasets. We evaluated the RNA binding feature map of 110 eCLIP-seq datasets and were able to recover several well-known features from the literature for RNA-binding proteins; we were also able to uncover novel associations. BioFeatureFinder is available at https://github.com/kbmlab/BioFeatureFinder/.