Pathway-Based Molecular Subtyping Reveals a GABA-Collapsed Autism Subgroup and Cross-Disease Convergence with Schizophrenia in Human Cerebral Cortex
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Autism spectrum disorder (ASD) and schizophrenia are genetically heterogeneous, with hundreds of risk genes contributing to single clinical diagnoses. This heterogeneity has frustrated molecular subtyping efforts and contributed to inconsistent drug trial outcomes. We present the Pathway Subtyping Framework, an open-source pipeline that shifts analysis from individual genes to biological pathways, and validate it on 444 postmortem brain transcriptome samples across three independent datasets spanning two diseases. In the discovery cohort (GSE28521, Voineagu et al., 2011; frontal cortex, n=32), pathway-based GMM clustering revealed a GABA-Collapsed subtype containing 56% of ASD samples and zero controls (Cohen's d = 3.21 for GABA signaling, p < 0.001). Independent validation on GSE64018 (Gupta et al., 2014; RNA-seq, temporal cortex, n=24) replicated the finding: a disease-enriched subtype (83% ASD) with the same top 3 disrupted pathways and overlapping driver genes (NRXN1, CNTNAP2, PTBP1), despite a different brain region and sequencing platform. Cross-disease analysis on GSE80655 (Ramaker et al., 2017; 141 schizophrenia + control samples, 3 brain regions) identified three molecular subtypes — Dopamine-Hyperactive (28%), Neurodevelopment-Activated (65%), and Synaptic-Collapsed (7%) — and was the first real-data analysis to pass all 3 validation gates (bootstrap ARI = 0.923). ASD and schizophrenia pathway sets produced 87% identical subtypes on the same data (ARI = 0.870). Multi-diagnosis pooled clustering of all 281 samples (schizophrenia, bipolar, depression, controls) showed that molecular subtypes are independent of clinical diagnosis (chi-squared p = 0.72) but strongly associated with brain region (p = 3 x 10^-16). These findings suggest that molecular subtypes cut across traditional diagnostic boundaries, with direct implications for precision drug selection in psychiatric clinical trials. The framework (v0.3.0, 968 tests) is available on PyPI ('pip install pathway-subtyping')