Whole genome sequencing is increasingly used to diagnose medical conditions of genetic origin. While both coding and non-coding DNA variants contribute to a wide range of diseases, most patients who receive a WGS-based diagnosis today harbour a protein-coding mutation. Functional interpretation and prioritization of non-coding variants represents a persistent challenge, and disease-causing non-coding variants remain largely unidentified. Depending on the disease, WGS fails to identify a candidate variant in 20–80% of patients, severely limiting the usefulness of sequencing for personalised medicine. Here we present FINSURF, a machine-learning approach to predict the functional impact of non-coding variants in regulatory regions. FINSURF outperforms state-of-the-art methods, owing in particular to optimized control variants selection during training. In addition to ranking candidate variants, FINSURF breaks down the score for each variant into contributions from individual annotations, facilitating the evaluation of their functional relevance. We applied FINSURF to a diverse set of 30 diseases with described causative non-coding mutations, and correctly identified the disease-causative non-coding variant within the ten top hits in 22 cases. FINSURF is implemented as an online server to as well as custom browser tracks, and provides a quick and efficient solution to prioritize candidate non-coding variants in realistic clinical settings.

Unraveling sequence determinants which drive protein-RNA interaction is crucial for studying binding mechanisms and the impact of genomic variants. While CLIP-seq allows for transcriptome-wide profiling of in vivo protein-RNA interactions, it is limited to expressed transcripts, requiring computational imputation of missing binding information. Existing classification-based methods predict binding with low resolution and depend on prior labeling of transcriptome regions to obtain high-quality training sets. We present RBPNet, a novel deep learning method, which predicts CLIP crosslink count distribution from RNA sequence at single-nucleotide resolution. RBPNet performs bias correction by modeling the raw CLIP-seq signal as a mixture of the (unobserved) protein-specific and background signal obtained from control experiments. By training on up to a million regions with elevated signal, RBPNet achieves better generalization over state-of-the-art classifiers on a variety of assays, including eCLIP, iCLIP and miCLIP. Through model interrogation via Integrated Gradients, RBPNet identifies highly predictive sub-sequences corresponding to known binding motifs and enables variant-impact scoring via in silico mutagenesis. Together, RBPNet improves inference of protein-RNA interaction, as well as mechanistic interpretation of predictions, by modeling the raw CLIP-seq data at high resolution.