Genome Clustering for Clinical Subtype Detection in Autism

University Of California, San Diego, La Jolla CA

Investigators

Linked publications & trials

Abstract

Project Summary/Abstract Autism spectrum disorder (ASD) is a developmental neuropsychiatric disorder characterized by a wide range of clinical manifestations and genetic heterogeneity. This complex genetic landscape poses a serious challenge to e?orts to ?nd a relationship between genotype and clinical phenotype in ASD. Therefore, the disorder remains poorly understand, and diagnostic and treatment tools are lacking. Though genetic overlap between patients is low, many mutations in autism likely a?ect stereotyped biological pathways, resulting in subgroups of autistic patients with similar functional mutation burden. The central hypothesis of this proposal is that patients with functionally similar mutations can be clustered together in order to reveal previously unidenti?ed groups that share both molecular mechanisms and disease related cognitive traits. This proposal ?rst describes an algorithm for genomic subtype detection of autistic patients using graph clustering and a novel distance function that is based on gene ontology, a database that describes the functional relationship of genes. Next, it is shown that resulting patient clusters are de?ned by both unique and convergent involvement of genetic pathways, which provides insight into the variety of disease mechanisms that may be present in ASD. These potential mechanisms can be explored at the cellular level by connecting the clustering algorithm with data from ASD patient derived neurons. Finally, cluster membership is associated with di?erences in the clinical and cognitive variables, validating the relevance of the genomic subtypes. In addition, deep learning approaches to describe individualized clinical prediction are discussed. Genome clustering in autism is an important step towards de?ning both mechanistic and clinical subtypes in this complex disorder, and this proposal addresses a growing need for algorithmic tools that connect genotype to phenotype in the context of heritable mental illnesses.