Design Of ASD Subtyping Approach based on Multi-Omics Data to Promote Personalized Healthcare

Abstract

Autism spectrum disorder (ASD) is a heterogeneous neurodevelopmental disorder that has been confirmed to be related to some genetics risk factors which can lead to different clinical phenotypes. At present, ASD is mainly diagnosed based on some behavior and cognitive scales, which can not reveal the mechanism of disease occurrence, development and prognosis. In recent years, some studies have applied omics techniques into ASD research, but these studies are only based on single omics data source such as genomics, proteomics or transcriptomic without investigating ASD subtypes from integration of multi-omics data. In this study, we proposed an ASD subtyping framework that integrates clinical and multi-omics data to identify and analyze ASD subtypes at the molecular level. Due to the heterogeneity of different data modalities, a fusion clustering strategy was used to produce more accurate and interpretable clusters. Based on ASD subtyping results, we also proposed a classification framework to predict the subtype of new ASD patients. Deep learning method was used to extract features from each data modality, then all extracted features were integrated by the multiple kernel learning method to improve the classification accuracy.