Deep embedded clustering by relevant scales and genome-wide association study in autism

Abstract Autism spectrum disorder (ASD) has heterogeneous phenotypic and genetic characteristics. Results from previous genome-wide association studies (GWAS) have identified numerous genetic variants associated with ASD. The association of these genetic variants with a single disease may be explained by a polygenic model in which the effects of each genetic variant are weak, yet contribute to the disease onset. When conducting GWAS with a certain sample size where no significant signal is produced, the signal will become increasingly difficult to identify if the sample size is reduced by dividing the patients. However, if ASD includes a subtype in which a small number of genes have a relatively strong influence, a subtype described by the oligogenic model, then it may be possible to identify some signals by dividing patients into phenotypically similar clusters and exploring genetic factors. To overcome ASD heterogeneity, we previously conducted cluster analyses of phenotypic variables from the Simons Simplex Collection dataset using k-means algorithm. In the present study, using a deep embedded clustering algorithm, we conducted cluster analyses of Simons Foundation Powering Autism Research for Knowledge (SPARK) datasets and performed GWAS of the clusters. We observed no significant associations in the conventional GWAS comparing all patients to all controls. However, in the GWAS, comparing patients divided into clusters with similar phenotypes to controls (cluster-based GWAS), we identified 90 chromosomal loci that satisfied the P < 5.0 × 10−8, several of which were located within or near previously reported candidate genes for ASD. Our findings suggest that ASD is an aggregation of etiologically heterogeneous subtypes, each of which has its own genetic architecture involving a smaller number of genetic loci than previously thought in addition to that explained by polygenic model.