Prediction of candidate genes for neuropsychiatric disorders using feature-based enrichment

The progress in understanding of molecular mechanisms underlying common heritable disorders (e.g. autism, schizophrenia, diabetes) depends on the availability of new bioinformatics approaches for identification of their characteristic genetic variations and associated multidimensional patterns of inheritance. High-throughput genome-wide studies (e.g. sequencing, gene expression profiling) result in hundreds of potential candidate genes. Prioritizing these genes and finding the best candidates contributing to a disease phenotype is one of the most important problems of genomics. We present an approach for prioritization of disease candidate genes using Support Vector Machine (SVM) and ontology associations. Features are extracted from both hierarchical and non-hierarchical ontology space (e.g user defined customized ontologies, Gene Ontology(GO) ). We select a subset of features according to enrichment scores in a training set of genes and use these to train a classifier using SVM. Ranking of the genes in the query set (e.g. the results of gene expression analysis) is based on a distance from the decision boundary to data points. Results obtained using the proposed approach to the analysis of several neurological disorders (autism, mental retardation, and agenesis of corpus callosum) are presented.