Artificial-intelligence- and multi-omics-guided predictive and drug repurposing framework construction for colorectal cancer: evidence from succinylation–neutrophil signatures

Abstract

Background Succinylation-linked metabolic rewiring and neutrophil-driven inflammation are key drivers of colorectal carcinoma (COAD) progression; however, they have rarely been translated to an end-to-end artificial intelligence (AI) drug-design pipeline connecting target nomination, resistance-relevant tumor microenvironment states, and candidate evaluation. Methods We integrated bulk transcriptomes from The Cancer Genome Atlas colon adenocarcinoma (TCGA-COAD) and Gene Expression Omnibus databases to derive succinylation–neutrophil (SN)-associated genes using the limma package, single-sample gene-set enrichment analysis, weighted gene coexpression network analysis, and protein–protein interaction network analysis; we then constructed a cross-cohort prognostic system by screening and selecting the machine-learning combination of random survival forest and ridge regression as the optimal predictor. To enable precision stratification, we applied deep-learning self-organizing maps (SOMs) to define the SN-driven molecular subtypes and characterized their immune and pathway heterogeneities at both bulk and single-cell resolutions through AI-powered virtual perturbation analysis to interrogate the cell-state shifts. Importantly, we extended AI approaches in drug design beyond risk modeling to a target-to-candidate workflow by nominating a druggable SN hub gene. We also performed co-culture in vitro assays for identifying NOX4 molecular insights into neutrophil patterns and their associations with COAD progression. Results Our approach identifies SN biology as a clinically actionable axis in COAD, highlights NOX4 as a central SN-associated target with immune relevance, and prioritizes talazoparib as a repurposing candidate supported by integrated in silico evidence. Conclusion The findings of this study provide a precision-oncology blueprint that couples AI stratification with therapeutic nomination and evaluation.