Conversational Artificial Intelligence-Based Integration of Clinical and Genomic Data Identifies MAPK Alterations in Colorectal Cancer

BackgroundColorectal cancer (CRC) exhibits marked heterogeneity across age, ancestry, and treatment context, underscored by the rising incidence of early-onset disease (EOCRC). The mitogen-activated protein kinase (MAPK) signaling axis is a central regulator of CRC biology and treatment response, yet the frequency, distribution, and prognostic relevance of MAPK pathway alterations across demographically and clinically defined subgroups remain unclear. MethodsWe analyzed 2,515 CRC tumors with harmonized clinical, genomic, and treatment metadata. Patients were stratified by ancestry (H/L vs. non-Hispanic White [NHW]), age at diagnosis (early-onset vs. late-onset), and FOLFOX chemotherapy exposure. Somatic MAPK alterations were identified using a curated gene set spanning canonical and regulatory components of EGFR-RAS-RAF-MEK-ERK signaling. Conversational artificial intelligence (AI-HOPE and AI-HOPE-MAPK) enabled cohort construction and exploratory analytics using natural language queries, with all findings validated using standard statistical methods. Overall survival (OS) associations were evaluated using Kaplan-Meier analyses. ResultsMAPK pathway alterations demonstrated marked heterogeneity across ancestry and treatment contexts. Among EO H/L patients, FGFR3, NF1, and RPS6KA6 mutations were significantly more frequent in those not treated with FOLFOX, while PDGFRB alterations were enriched in EO H/L patients receiving FOLFOX compared with EO NHW counterparts. In late-onset H/L disease, NTRK2 and PDGFRB mutations were more prevalent in non-FOLFOX-treated tumors. Distinct MAPK-related alterations were also observed among NHW patients, including enrichment of AKT3, FGF4, RRAS2, CRKL, DUSP4, JUN, MAPK1, RRAS, and SOS1 mutations in non-FOLFOX-treated groups, as well as age-dependent differences in BRAF and TP53 mutation frequencies. Survival analyses revealed borderline evidence that MAPK pathway alterations were associated with improved overall survival in select NHW subgroups, including EOCRC patients treated with FOLFOX. Conversational AI enabled rapid generation and validation of multi-parameter queries. ConclusionsAlthough MAPK pathway disruption is a near-universal molecular feature in CRC, the specific genes, effectors, and regulatory nodes altered vary by ancestry, age of onset, and chemotherapy exposure, revealing biologically and clinically meaningful substructure that is not captured by pathway-level status. NF1, MAPK3, RPS6KA4, and PDGFRB emerge as candidate genomic biomarkers in EOCRC and in H/L populations, while BRAF and adaptor-mediated signaling alterations are enriched in late-onset CRC. Conversational AI significantly accelerated analytic throughput and supported equity-driven biomarker discovery. Future studies incorporating multi-omic data and diverse prospective cohorts will be essential to refine MAPK-based precision strategies and address population-level disproportions in CRC.