@article{BHAVNA2026133145,
title = {ACVAE: An attention-based contrastive variational autoencoder for identifying ASD subgroups and their association with gene expression},
journal = {Neurocomputing},
volume = {678},
pages = {133145},
year = {2026},
issn = {0925-2312},
doi = {https://doi.org/10.1016/j.neucom.2026.133145},
url = {https://www.sciencedirect.com/science/article/pii/S0925231226005424},
author = {Km Bhavna and Shubham Sharma and Alfredo Rosado-Muñoz},
keywords = {ASD sub-groups, Spatio-temporal connectivity, Attention-based contrastive variational autoencoder model, Gene expression},
abstract = {Autism Spectrum Disorder (ASD) is characterized by deficits in social cognition, interaction, communication, restricted behaviours, and sensory abnormalities. Functional MRI (fMRI) studies have implicated the Theory-of-Mind (ToM) Network, Default-Mode Network (DMN), Central Executive Network (CEN), and Salience Network (SN) in ASD. The disorder’s diverse clinical presentation complicates diagnosis and treatment, necessitating a deeper understanding of its heterogeneity. This study aims to identify novel features from resting- state fMRI data to identify sub-groups within ASD samples. We proposed an Attention-based Contrastive Variational Autoencoder (ACVAE) model to identify ASD subgroups using higher- order edge connectivity (spatio-temporal connectivity) among ToM, DMN, CEN, and SN networks. Using the Autism Brain Imaging Data Exchange (ABIDE) dataset, we analyze static, dynamic, and spatiotemporal functional connectivity to understand brain topology. Our method successfully identified three ASD subgroups and their associations with gene expression. We integrate functional connectivity matrices with gene expression profiles to explore biological underpinnings, examining covariation between fMRI-derived patterns and key genes linked to ASD-related functional disruptions. This integrative approach bridges neuroimaging and genomics, offering insights into ASD heterogeneity and potential biomarkers. To our knowledge, this is the first study using ACVAE with higher-order connectivity features and gene expression integration to uncover ASD subgroups.}
}