Accelerate the discovery of genetic variants in mitochondrial diseases with VIOLA: Variant PrIOritization using Latent space

Abstract

Interpreting variants from whole-exome sequencing remains a major challenge, particularly for heterogeneous disorders such as mitochondrial diseases (MD). To help finding a diagnosis for complex cases, we have developed VIOLA (Variant prIoritizatiOn using Latent spAce). The pipeline includes a variational autoencoder to embed functional annotations into a low-dimensional space, followed by DBSCAN-based outlier detection. To prioritize potential pathogenic variants, filtering steps and phenotype integration via HPO terms are then applied. Two scores are associated to the selected variants: the VIOLA score (Vscore), which combines variant features, transcriptomic co-expression data, and MD-specific annotations, and the VIOLA Aggregated score (VAscore) that merges Vscore with Exomisers pathogenicity score. Finally, we also provide the ARrank, specifically designed for variants compatible with autosomal recessive inheritance. We illustrate the functionalities of VIOLA on a novel cohort of patients with suspected MD with complex phenotypes. VIOLA systematically ranked causal variants among the top, outperforming existing methods. Overall, VIOLA is a patient-specific approach to help to discover novel variants in complex MD.