Artificial intelligence–enabled precision oncology in early-onset colorectal cancer: Molecular characterization using spatial proteomics, spatial transcriptomics, multi-omics, microbiome, social determinants, and clinical data.

Abstract

192 Background: Early-onset colorectal cancer (EOCRC; <50 years) is increasing at an alarming rate and disproportionately affects Hispanic/Latino (H/L) and other underserved populations. The biological drivers of EOCRC and the influence of clinical exposures and social determinants of health (SDoH) remain poorly defined. Spatial proteomic and transcriptomic technologies enable high-resolution analysis of tumor, immune, and stromal compartments, while artificial intelligence (AI) offers opportunities for data integration. We hypothesized that EOCRC in H/L patients harbors ancestry-associated alterations and spatial expression patterns that influence treatment response and outcomes. Methods: We conducted a retrospective, observational multi-cohort study of 4,874 CRC patients (early- and late-onset, multi-ethnic) using institutional and public datasets. Data included spatial proteomics (Phenocycler 2.0), spatial transcriptomics (10x Genomics Visium HD), RNA sequencing, whole-exome sequencing (WES), and microbiome (16S/shotgun) profiling. Clinical records and SDoH variables were integrated to evaluate ancestry-specific associations with different therapies (FOLFOX, FOLFIRI, anti-EGFR, anti-VEGF, immunotherapy). Standard pipelines were applied, followed by Artificial Intelligence-enabled integration using conversational agents for cross-domain exploration of molecular and clinical features. Results: WES revealed ancestry-linked molecular profiles distinguishing EOCRC from late-onset CRC, particularly in H/L patients. Integrated WES/RNAseq analyses confirmed ancestry-specific oncogenic pathway alterations. Spatial proteomics and transcriptomics demonstrated marked compartmental heterogeneity, with EOCRC tumors displaying distinct pathway activation in epithelial, immune, and stromal niches. Multi-omics and microbiome profiling highlighted enrichment of pro-tumorigenic taxa and differential pathway activity in EOCRC. Incorporation of clinical and SDoH variables revealed ancestry-associated differences in treatment response, with preliminary analyses showing distinct overall survival patterns between EOCRC and late-onset CRC under chemotherapy, targeted therapy, and immunotherapy. Conclusions: AI-enabled molecular characterization of EOCRC reveals ancestry-associated alterations, spatial heterogeneity, and treatment-linked disparities, particularly in H/L populations. Integrating spatial, multi-omic, microbiome, clinical, and SDoH data establishes a comprehensive framework for precision oncology in EOCRC. These findings emphasize the importance of ancestry and SDoH in biomarker discovery and treatment optimization, advancing precision medicine for young patients.