WNT pathway alterations in bevacizumab-treated early-onset colorectal cancer: An artificial intelligence–driven and disparity-aware precision oncology analysis.

235 Background: The WNT signaling pathway, dominated by APC and related regulators, is central to colorectal cancer (CRC), yet its prognostic and predictive relevance under anti-angiogenic therapy is unclear. Early-onset CRC (EOCRC), rising most rapidly among Hispanic/Latino (H/L) populations, may show context- and ancestry-specific WNT biology. We evaluated WNT alterations as candidate biomarkers for outcome stratification in bevacizumab-treated CRC. Methods: We retrospectively analyzed 2,717 CRC patients across public datasets with genomic, clinical, and treatment annotations. Eligible cases had colorectal adenocarcinoma, sequencing data, and bevacizumab exposure status. Tumors included primary and metastatic samples. Subgroups were defined by age (<50 years EOCRC vs. ≥50 years LOCRC) and ancestry (H/L vs. non-Hispanic White NHW). The primary endpoint was overall survival (OS); secondary endpoints were mutation frequency differences and cross-strata comparisons. WNT alterations (APC, RNF43, AXIN1/2, TCF7L2, AMER1) were analyzed as categorical variables. Frequencies were compared using Fisher’s exact test; OS was assessed by Kaplan–Meier and Cox models. Analyses were exploratory and unadjusted for multiple comparisons. Conversational AI agents enabled reproducible multi-parameter queries. Results: WNT alterations were frequent and dominated by APC. In H/L EOCRC, bevacizumab-exposed tumors had lower RNF43 frequency (0% vs. 14.3%, p = 0.03), with similar findings in H/L LOCRC (2.3% vs. 18.0%, p = 0.008). In NHW EOCRC, bevacizumab exposure corresponded to reduced RNF43 (1.7% vs. 7.7%, p = 0.003), TCF7L2 (11.6% vs. 20.9%, p = 0.008), AXIN1 (0% vs. 2.6%, p = 0.02), and AXIN2 (1.2% vs. 5.1%, p = 0.02). In NHW LOCRC, exposed tumors showed lower AMER1 (5.9% vs. 11.0%, p = 0.005), AXIN1 (0.8% vs. 4.7%, p = 0.0001), AXIN2 (3.8% vs. 8.3%, p = 0.004), RNF43 (4.9% vs. 13.8%, p = 4.05×10⁻⁶), and TCF7L2 (11.9% vs. 16.9%, p = 0.01). Cross-age comparisons in NHW unexposed cohorts showed higher APC and TCF7L2, but lower AXIN1, AXIN2, and RNF43 in EOCRC relative to LOCRC. Among unexposed LOCRC, H/L tumors had lower APC than NHW (65.8% vs. 75.1%, p = 0.03). Survival analyses revealed WNT alterations predicted improved OS in H/L EOCRC (p = 0.014), worse OS in NHW EOCRC (p = 0.023), and improved OS in NHW LOCRC (p = 0.0052). Conclusions: This exploratory biomarker study identifies WNT alterations as context-dependent prognostic markers in CRC, with outcomes modified by treatment, ancestry, and age. Associations between bevacizumab and reduced WNT mutation frequencies suggest potential interaction between anti-angiogenic therapy and WNT signaling. Prospective validation with standardized treatment and integration of tumor microenvironment and social determinants is warranted to advance equitable precision oncology.