Genome-Wide Uncertainty-Moderated Extraction of Signal Annotations from Multi-Sample Functional Genomics Data
Multi-sample functional genomics experiments should reveal reproducible regulatory activity, but locus- and sample-specific noise can obscure biological signals in sequencing data. We introduce Consenrich for genome-wide estimation of epigenomic signals across multiple samples. To encourage robustness and sensitivity, the state-space model underlying Consenrich accounts for positional observation variances to determine shrinkage toward predictions from a smooth process model over genomic coordinates. We first apply Consenrich to ATAC-seq and ChIP-seq datasets and demonstrate its ability for robust signal recovery. We then utilize Consenrich upstream of a class-imbalanced differential accessibility analysis in an Alzheimer's cohort of twenty samples and show that it improves the breadth of relevant biological insights. Software is available at https://github.com/nolan-h-hamilton/Consenrich.