Elucidating disease mechanisms via their tissue-selectivity

Due to the large number of genetic variants in our genomes whose functional significance remains unknown, the general rate of successful genetic diagnosis of Mendelian diseases is ~25-50%. One of the hallmarks of Mendelian diseases is their tendency to manifest clinically only in few tissues, making them highly tissue-selective. Multiple studies by others and by us have uncovered common transcriptome-based and interactome-based features that signify causal genes in their affected tissues. However, tissue-specific 'omics data have not been routinely considered when genetically diagnosing rare diseases. Here we present a machine-learning scheme that uses thousands of omics-based features for prioritizing candidate genes in 8 different human tissues. Genes that are known to be causal for a tissue-specific disease ranked significantly higher in that tissue relative to non-causal genes, and, notably, also relative to genes causal for diseases manifesting in other tissues. Promising results were also obtained upon applying our scheme to data from patients. Thus, machine learning of tissue-specific 'omics can boost current genetic diagnosis schemes.