Non-apoptotic regulated cell death based prognostic risk model for colorectal cancer using machine learning guided...

Colorectal cancer (CRC) prognosis is severely limited by tumor heterogeneity. To address this, we leveraged the emerging role of non-apoptotic regulated cell death (NARCD) pathways to develop a two-step machine learning (ML) framework to develop a prognostic risk model. We used two largest and independent CRC patient cohorts (TCGA and E-MTAB-12862). Our novel pipeline consists of two steps, first, a library of 46 combination ML survival models was applied to each of the 13 NARCD pathways to establish robust, pathway-specific prognostic models. Second, logistic regression was used to integrate these models, revealing that the synergistic combination of ferroptosis, NETosis, pyroptosis, and autosis yielded the highest predictive power. The resulting 43-gene prognostic risk model, the combined-regulated cell death index (c-RCDI), robustly stratified patients and proved to be a powerful independent prognostic factor (HR = 55.1, P < 0.001), with high predictive accuracy (5-year AUROC = 0.88). Notably, this prognostic power was exclusive to CRC. Biologically, the high-risk class showed enriched angiogenesis and EMT pathways, an immunosuppressive microenvironment, reduced immunotherapy response, and predicted increased sensitivity to CDK, MEK, and metabolic inhibitors, while the low-risk class showed increased sensitivity to the drug "Obatoclax Mesylate_1068". c-RCDI developed in this study using a novel two-step ML framework based on NARCD pathways showed robust predictive ability for CRC patients, with a potential for improving diagnosis and therapy.