Feature Selection for Autism Spectrum Disorder via a Multi-Pack Cooperative Grey Wolf Optimization Framework
TL;DR
The identified gene signature showed improved classification accuracy, and its enriched biological functions provided insights into ASD-related molecular mechanisms, suggesting potential value for future ASD-related genomic research and non-invasive diagnostic exploration.
Abstract
Autism Spectrum Disorder (ASD) is a prevalent neurodevelopmental condition in childhood and adolescence. However, its heterogeneous genetic mechanisms make early diagnosis particularly challenging. Identifying robust molecular biomarkers from high-dimensional gene expression data remains a critical bottleneck. In this study, we proposed the Multi-pack cooperative grey wolf optimization (MPC-GWO) algorithm to select robust gene biomarkers for ASD using two public blood-based transcriptomic datasets (GSE25507 for training, GSE18123 for independent testing). MPC-GWO was applied to GSE25507 to identify the minimal gene subset that reliably discriminated ASD from controls. The selected biomarker genes were then evaluated on GSE18123 to test cross-dataset generalizability. After statistical preselection and MPC-GWO refinement, we identified a 24-gene signature that achieved superior classification performance on the discovery cohort (accuracy=0.808, AUC=0.823), outperforming both the all-features baseline (accuracy=0.713) and filter-based t-test selection (accuracy=0.678). Compared with LASSO (accuracy=0.732, AUC=0.756) and Random Forest (accuracy=0.678, AUC=0.737), MPC-GWO demonstrated superior classification performance. The algorithm reduced the feature set by >88% and converged within 50 iterations. On the independent cohort, the signature achieved accuracy=0.674 and AUC=0.733. Functional enrichment analysis revealed that the selected genes are strongly associated with nervous system development, Ig-like C2-type, neurodevelopmental disorders, and extracellular space, most of which have been repeatedly implicated in ASD. These results demonstrate that MPC-GWO is effective for ASD biomarker discovery. The identified gene signature showed improved classification accuracy, and its enriched biological functions provided insights into ASD-related molecular mechanisms, suggesting potential value for future ASD-related genomic research and non-invasive diagnostic exploration.