Application of Hi-C and machine-learning tools towards detection and clinical interpretation of novel genomic variants in rare genetic diseases

Abstract

Most rare diseases have an underlying genetic cause for their condition. There are more than 7,000 rare diseases, ~80% of which are thought to have a genetic cause. The majority (50-75%) of rare diseases affect children, and many are severe multisystem disorders with a range of phenotypes. Although individually they are rare, collectively they are responsible for 35% of deaths in the first year of life and are a significant cause of pediatric hospital admissions; one-third of children born with a rare disease will not live to see their fifth birthday.Accurate discerning the precise molecular cause (genotype) that explains the clinical features of the rare disease (phenotype) is the cornerstone of safe medical practice. For children with a rare genetic disorder, a robust genetic diagnosis unlocks access to a wealth of information in the literature that provides advice on management and therapy and enables access to disorder-specific support groups, which reduces isolation for families affected by a rare disorder. However, finding a molecular diagnosis remains a considerable challenge because of the genetic and phenotypic variability associated with these diseases and our incomplete knowledge.There are two major limitations of current methods of genetic diagnosis. First, existing methods vary in their sensitivity and resolution of analysis, and none of them can detect both copy-number variants (CNVs), structural variations (SV), and single nucleotide variants (SNVs) with high recall and precision. Sensitivity of the available methods is especially limited in case of the balanced submicroscopic SVs and complex SVs involving multiple chromosomal segments. Recently, derivates of the Hi-C approach gain attention as powerful tools for detection of complex chromosomal rearrangements.The second challenge is the accurate classification and assessing causality of non-codingvariants. Whereas there are multiple databases and tools allowing to estimate effects of protein-coding variants, clinical interpretation of non-coding SNVs and SVs is yet uncertain. Recent experimental and in silicomethods allowing profiling of the transcriptome, epigenome, and 3D-organization of chromatin opened new possibilities for functional analysis of non-coding variations. In this Project, we propose to apply experimental and computational techniques established by Veniamin Fishman (visiting scientist) and his group in Novosibirsk to contribute to improve the existing methods for molecular diagnosis in the Human Genome and Stem Cell Research Center. In particular, the main contribution of this collaboration to CEPID is the proposal of using the Hi-C and other 3C-based methods to resolve SVs with high resolution and the use modern AI-based tools to assess effects of non-coding genomic variants on gene expression and splicing. (AU)