Conversational AI agent for precision oncology: AI-HOPE-WNT integrates clinical and genomic data to investigate WNT pathway dysregulation in colorectal cancer.

INTRODUCTION

The WNT signaling pathway is a key driver of colorectal cancer (CRC) initiation and progression, particularly in early-onset CRC (EOCRC) among underserved populations. However, interrogating WNT pathway dysregulation across clinical and genomic dimensions remains technically challenging, limiting both translational insight and personalized intervention strategies. To address this gap, we developed AI-HOPE-WNT, the first conversational artificial intelligence (AI) agent purpose-built to investigate WNT signaling in CRC using natural language-driven, integrative bioinformatics.

METHODS

AI-HOPE-WNT employs a modular architecture combining large language models (LLMs), a natural language-to-code engine, and a backend statistical workflow interfaced with harmonized data from cBioPortal. Unlike general-purpose platforms, AI-HOPE-WNT is uniquely optimized for WNT-specific precision oncology. The tool supports mutation frequency analysis, odds ratio testing, survival modeling, and subgroup stratification by genomic, clinical, and demographic variables. To validate the platform, we recapitulated findings from two previous studies examining WNT pathway alterations in high-risk CRC populations, including mutation prevalence in RNF43 and AXIN2 and survival outcomes associated with WNT pathway status across ethnic and age subgroups. Exploratory queries further assessed treatment response, co-mutation patterns, and population-specific trends.

RESULTS

In recapitulation analyses, AI-HOPE-WNT reproduced key trends from prior work, including improved survival in WNT-altered EOCRC and higher RNF43 mutation rates in Hispanic/Latino (H/L) populations compared to non-Hispanic White (NHW) people. Exploratory analyses revealed several novel findings. Among FOLFOX-treated EOCRC patients, APC mutations were associated with significantly different survival outcomes ( = 0.043). RNF43-mutant tumors showed worse survival in metastatic versus primary cases ( = 0.028). AXIN1 and APC co-mutations demonstrated location-specific enrichment between colon and rectal tumors. Gender-based differences in AXIN2-mutant cases under varying MSI status yielded significant survival variation ( = 0.036). Additionally, patients under 50 with APC-mutant primary tumors showed worse survival ( = 0.031) and increased mutation prevalence.

CONCLUSION

AI-HOPE-WNT is the first dedicated AI platform for WNT pathway analysis in CRC. By combining natural language interaction with automated, high-throughput bioinformatics, it democratizes access to pathway-specific precision oncology research. The platform is freely available at: https://github.com/Velazquez-Villarreal-Lab/AI-HOPE-WNT.