Shared Molecular Signatures and Pathways in Post-Traumatic Stress Disorder, Schizophrenia, and Bipolar Disorder: An Integrative Transcriptomic and Machine-Learning Study

Abstract Objective: Psychiatric disorders such as post-traumatic stress disorder (PTSD), schizophrenia, and bipolar disorder (BD) are complex conditions with overlapping molecular features that remain poorly understood. This study aimed to identify shared molecular abnormalities and potential therapeutic targets associated with these disorders. Methods: RStudio (version 4.4.1) was used to process and analyze gene expression datasets from the NCBI database to identify differentially expressed genes (DEGs). FunRich, DAVID, STRING, and NetworkAnalyst were used for gene ontology and pathway enrichment, protein-protein interaction analysis, transcription factor and microRNA regulatory network analysis, and protein-chemical interaction analysis. The discriminatory ability of prioritized hub genes and transcription factors was evaluated using receiver operating characteristic (ROC) analysis. Four machine-learning algorithms (Logistic Regression, Random Forest, SVM-RBF, and Gradient Boosting) were compared using leave-one-out cross-validation on the GSE403628 dataset, comprising 14 controls and 14 cases. Results: Among 32 common genes, several candidate hub proteins were identified, including HLA-DRA, HLA-A, HLA-B, HLA-DOB, and BRD2. The analysis also identified transcription factors (FOXC1, NFYA, RELA, GATA2, FOXL1, SRF, and NFIC), miRNAs (hsa-mir-129-2-3p, hsa-mir-148b-3p, hsa-mir-196a-5p, hsa-mir-26a-5p, hsa-mir-27a-3p, hsa-mir-23b-3p, hsa-mir-500a-3p, hsa-mir-423-5p, hsa-miR-142-5p, and hsa-miR-671-5p), and chemicals, including estradiol, antirheumatic agents, and valproic acid. ROC analysis showed the highest AUC for HLA-DRA (0.760) among hub genes and FOXC1 (0.658) among transcription factors. The hub-gene set consistently outperformed the transcription-factor set across all four machine-learning algorithms, with Logistic Regression achieving the highest AUC (0.800). Random Forest and SVM-RBF achieved the highest accuracy (0.679) under leave-one-out cross-validation. Conclusion: This computational approach identified potential biomarkers linking PTSD, schizophrenia, and BD to immune-related and infectious disease pathways. ROC and machine-learning analyses highlighted HLA-DRA as a promising discriminator between disease and control samples, providing candidate molecular targets for further investigation.