Integrated machine learning and causal inference reveal CHRDL1 as a diagnostic and functional biomarker in NSCLC

Non-small cell lung cancer (NSCLC) accounts for the majority of lung cancer-related deaths, largely due to late diagnosis and limited early biomarkers. Integrating machine learning (ML) with causal inference may improve biomarker discovery and translational potential.

Six NSCLC transcriptomic datasets were integrated and normalized. Differentially expressed genes were intersected with WGCNA modules to identify candidate features. A total of 114 ML models were constructed using various feature selection and classification algorithms, with model performance evaluated by area under the ROC curve (AUC). Immune cell infiltration was assessed using CIBERSORT. Causal relationships were evaluated via two-sample Mendelian randomization (MR) using pQTL and GWAS data. Functional pathways were explored using Gene Set Enrichment Analysis (GSEA) and Gene Set Variation Analysis (GSVA).

A Random Forest model achieved the best performance (average AUC = 0.994) and identified nine diagnostic genes, including CHRDL1, FCN3, WIF1, MMP12, and AKR1B10. These genes showed strong diagnostic power (AUC > 0.80) and distinct immune-related expression patterns. MR analysis supported a causal protective role for CHRDL1 in NSCLC (IVW p = 0.0012), with no evidence of pleiotropy or heterogeneity. GSEA and GSVA revealed that high CHRDL1 expression was associated with immune activation, while low expression correlated with cell cycle progression.

This integrative ML-MR framework identified CHRDL1 as a robust diagnostic and functionally relevant biomarker for NSCLC. These findings offer insights into the immune-proliferative landscape of NSCLC and provide a foundation for early detection and targeted therapy strategies.