Multimodal-based deep learning detected disrupted precuneus connectivity and its related genetic profiles for predicting adults with ADHD.
Authors
Affiliations (6)
Affiliations (6)
- Peking University Sixth Hospital, Institute of Mental Health, National Clinical Research Center for Mental Disorders (Peking University Sixth Hospital), China; Key Laboratory of Mental Health, Ministry of Health, Peking University, Beijing, 100191, China.
- Department of Medical Genetics, School of Basic Medical Sciences, Peking University, Beijing, China.
- Peking University Sixth Hospital, Institute of Mental Health, National Clinical Research Center for Mental Disorders (Peking University Sixth Hospital), China; Key Laboratory of Mental Health, Ministry of Health, Peking University, Beijing, 100191, China; School of Psychological Science, University of Western Australia, Perth, Australia.
- Peking University Sixth Hospital, Institute of Mental Health, National Clinical Research Center for Mental Disorders (Peking University Sixth Hospital), China; Key Laboratory of Mental Health, Ministry of Health, Peking University, Beijing, 100191, China. Electronic address: [email protected].
- Department of Medical Genetics, School of Basic Medical Sciences, Peking University, Beijing, China; Center for Medical Genetics, School of Basic Medical Sciences, Peking University, Beijing, China; Medicine Innovation Center for Fundamental Research on Major Immunology-related Diseases, Peking University, Beijing, China; Department of Biomedical Informatics, School of Basic Medical Sciences, Peking University, Beijing, China. Electronic address: [email protected].
- Peking University Sixth Hospital, Institute of Mental Health, National Clinical Research Center for Mental Disorders (Peking University Sixth Hospital), China; Key Laboratory of Mental Health, Ministry of Health, Peking University, Beijing, 100191, China. Electronic address: [email protected].
Abstract
Integrating neuroimaging genetics data may enhance the diagnosis and understanding of adults with attention-deficit/hyperactivity disorder (ADHD). Here, we employed a population-based graph convolutional network (GCN) model based on multimodal data for predicting adults with ADHD, and performed downstream analyses to elucidate the related pathophysiology. Functional MRI and genomics data were collected from 258 adults with ADHD and 243 controls. Multimodal data were fused and trained via an Edge-Variational Graph Convolution Network. Through sensitivity analysis, functional connectivity (FC) and single nucleotide polymorphisms contributing to ADHD were identified and validated by correlating them with clinical manifestations. Downstream association analyses, which included GWAS and transcriptome-neuroimaging association analyses, were conducted to explore the genetic components linked to imaging alterations. Our model achieved an accuracy of 73.55 %, outperforming the other prediction methods. Among the most salient FC, FC between the bilateral precuneus (PCun) and left parahippocampal gyrus as well as right hippocampus correlated significantly with ADHD core symptoms. Key hubs particularly the left PCun were identified from the salient FC, whose topological properties are associated with hyperactivity/impulsivity and inhibitory control deficits. Notably, deficient PCun-based FC exhibited relations with genetic variants of KDM1A and HSPG2, and with expression profiles of MDGA1. These findings highlight the effectiveness and interpretability of the multimodality-based deep learning in discriminating adults with ADHD. Functional alterations in PCun and related genetic profiles may play an important role in adults with ADHD.