Machine learning approach reveals novel pathways in ASD and Schizophrenia

Autism spectrum disorder (ASD) and schizophrenia are highly heritable neurodevelopmental disorders. Common genetic variants contribute substantially to ASD and schizophrenia susceptibility, however, most direct connections between gene function and pathogenesis of these disorders have been provided by rare variation. For diagnostics purposes, connections between common variants and pathology should be made. To establish such connections we reanalyse published datasets of case-control studies containing 4026 ASD patients and 6135 schizophrenia patients and similar number of controls. We use WizWhy, a data mining tool that uses an association rules algorithm for supervised learning, to analyse 1077605 common missense SNPs in 18498 genes. We first identify rules that have disease probability distinct from the cohort disease probability. Then we test that these rules hold in two different subsets of cases and controls that were not used in the first stage. Enrichment analysis of the of genes involved in the top 1000 rules highlights known functions impaired in ASD and schizophrenia such as synapse formation, as well as novel functions such as sensory modulation that fit the phenotype of ASD and schizophrenia patients.