Integrated Metaheuristic-Deep Learning Framework for Cross-Disorder Genetic Analysis of Schizophrenia and Major Depressive Disorder

Abstract This comprehensive study presents an integrated metaheuristic-deep learning framework for cross-disorder analysis of schizophrenia (SCZ) and major depressive disorder (MDD) using genome-wide association studies (GWAS) data. Our methodology combines: (1) a multi-objective genetic algorithm with biological constraints for SNP selection, (2) a dual-task convolutional neural network for joint disorder prediction, and (3) multi-modal biological validation. The optimized model achieves 93.7% accuracy for SCZ (AUC=0.96) and 90.2% for MDD (AUC=0.93) with 300 training epochs, while identifying 1,842 shared risk SNPs showing significant enrichment in calcium signaling pathways (\(p=3.2 × 10⁻¹⁴\)). These findings establish a scalable framework for cross-disorder genetic analysis in psychiatry with detailed mathematical formulations of optimization objectives and network architecture specifications.