Artificial intelligence for precision oncology: AI-HOPE-MAPK uncovers clinically actionable MAPK alterations in colorectal cancer

Abstract

BackgroundThe emergence of early-onset colorectal cancer (EOCRC), particularly among populations with disproportionate health burdens, has exposed critical gaps in our understanding of age- and ancestry-specific oncogenic processes. Aberrations in the mitogen-activated protein kinase (MAPK) pathway--encompassing KRAS, NRAS, BRAF, MAP2K1, MAPK3, and others--are central to tumorigenesis, yet their distribution, prognostic impact, and therapeutic implications in EOCRC remain poorly defined. MethodsWe developed AI-HOPE-MAPK, a conversational artificial intelligence platform built upon a fine-tuned biomedical large language model (LLaMA 3) to enable real-time, natural language-driven analysis of clinical and genomic data. The platform integrates harmonized datasets from TCGA, cBioPortal, and AACR Project GENIE and supports cohort construction, mutation frequency comparisons, survival modeling, and odds ratio estimation across diverse subgroups. We applied AI-HOPE-MAPK to characterize MAPK pathway alterations in colorectal cancer by age, race/ethnicity, microsatellite instability (MSI), tumor mutational burden (TMB), treatment history, and anatomical site. ResultsAI-HOPE-MAPK identified significant age- and ancestry-specific patterns in MAPK alterations. NF1 and MAP2K1 mutations were enriched in H/L EOCRC cases, with NF1 mutations reaching statistical significance (p = 0.045). Comparative analyses revealed elevated MAPK3 and NF1 mutation frequencies in Hispanic/Latino and non-Hispanic White EOCRC patients. MAP2K1-mutated stage I-III tumors exhibited worse survival and lower FOLFOX treatment rates (p = 0.0447 and p = 0.015, respectively). Kaplan-Meier and odds ratio analyses showed that younger patients with low-TMB MAPK-altered tumors had significantly poorer outcomes and were overrepresented in aggressive molecular subtypes. Furthermore, MAPK alterations conferred significantly reduced survival in microsatellite-stable (MSS) tumors but not in MSI-H cases, underscoring context-specific prognostic effects. Anatomical stratification showed site-specific impacts of AKT1 mutations, with poorer outcomes in colon versus rectal cancers. ConclusionsAI-HOPE-MAPK offers a transformative framework for AI-driven precision oncology by enabling intuitive, scalable, and population-aware exploration of MAPK pathway dysregulation in colorectal cancer. The platform revealed clinically actionable differences in mutation prevalence, survival outcomes, and treatment access, particularly among patients with disproportionate health burdens. By combining language-based user interfaces with rigorous genomic analytics, AI-HOPE-MAPK accelerates biomarker discovery, supports equitable research practices, and advances the implementation of precision oncology across multiple patient populations.