Artificial Intelligence-Enhanced Molecular Profiling of JAK-STAT Pathway Alterations in FOLFOX-Treated Early-Onset...

Early-onset colorectal cancer (EOCRC) continues to rise, with the steepest increases observed among Hispanic/Latino (H/L) populations, underscoring the urgency of identifying ancestry- and treatment-specific biomarkers. The JAK-STAT signaling axis plays a central role in colorectal tumor biology, yet its relevance under FOLFOX-based chemotherapy in EOCRC remains poorly defined. In this study, we evaluated 2515 colorectal cancer (CRC) cases (266 H/L; 2249 non-Hispanic White [NHW]), stratifying analyses by ancestry, age of onset, and FOLFOX exposure. Statistical comparisons were performed using Fisher's exact and chi-square tests, and survival patterns were assessed via Kaplan-Meier analysis. To extend conventional analytics, we deployed AI-HOPE and AI-HOPE-JAK-STAT, conversational artificial intelligence platforms capable of harmonizing genomic, clinical, demographic, and treatment variables through natural language queries, to accelerate multi-parameter biomarker exploration. JAK-STAT pathway alterations showed marked variation by ancestry and treatment context. Among H/L EOCRC cases, alterations were significantly enriched in patients who did not receive FOLFOX compared with those who did (21.2% vs. 4.1%; p = 0.003). A similar pattern emerged in late-onset CRC (LOCRC) NHW patients, where alterations were more frequent without FOLFOX exposure (13.3% vs. 7.5%; p = 0.0002). Notably, JAK-STAT alterations were significantly more common in untreated H/L EOCRC compared with untreated NHW EOCRC (21.2% vs. 9.9%; p = 0.002). Survival analyses revealed that JAK-STAT pathway alterations conferred improved overall survival across several NHW strata, including EOCRC treated with FOLFOX (p = 0.0008), EOCRC not treated with FOLFOX (p = 0.07), and LOCRC not treated with FOLFOX (p = 0.01). These findings suggest that JAK-STAT alterations may function as ancestry- and treatment-dependent prognostic markers in EOCRC, particularly among disproportionately affected H/L patients. However, prognostic interpretation in H/L subgroups is limited by small mutation-positive sample sizes, reflecting historical underrepresentation and highlighting the need for larger ancestry-balanced studies. The integration of AI-enabled platforms streamlined analyses and reveals the potential of artificial intelligence to accelerate discovery and advance precision medicine for populations historically underrepresented in cancer genomics research.