Three AI-Enabled Analyses Highlight Context-Dependent Biomarkers in Early-Onset Colorectal Cancer

Biomarker discovery in colorectal cancer has traditionally focused on identifying molecular alterations with broad prognostic or predictive utility. However, evidence is increasingly suggesting that biomarkers do not have universal prognostic or predictive value across patient sets but instead depend on contextual factors such as ancestry, age at onset, and treatment exposure.

Three posters presented at the 2026 ASCO Gastrointestinal Cancers Symposium, all led by Enrique Velazquez Villarreal, MD, PhD, MPH, MS, Assistant Professor, Department of Integrative Translational Sciences, City of Hope, Duarte, California, offer complementary analyses of artificial intelligence (AI)–backed biomarker discovery in early-onset colorectal cancer. In each study, researchers from Velazquez Villarreal’s lab used AI tools to simplify complex, multidimensional datasets to address clinical and translational research questions.

Together, the studies highlight a shift in biomarker development toward models that account for clinical and sociodemographic contexts. They also demonstrate how AI can support scalable, reproducible analyses of complex molecular and clinical data to inform precision oncology approaches in higher-risk populations, including patients from racial and ethnic minority groups.

Multiomics for Precision Oncology

The researchers hypothesized in the first poster that treatment responses and outcomes in Hispanic/Latino patients with early-onset colorectal cancer may be driven by ancestry-influenced molecular alterations and spatial expression patterns (Abstract 192). They conducted a retrospective, observational multicohort study of 4,874 patients with colorectal cancer, including both those with early- and late-onset disease. They gathered data from spatial proteomics, spatial transcriptomics, RNA sequencing, whole-exome sequencing, microbiome profiling, and social determinants of health. With the use of AI, all the analyses were integrated with conservation agents to explore molecular and clinical features of disease.

Ancestry-linked molecular profiles in oncogenic pathways were more frequently observed in patients with early-onset colorectal cancer compared with those with late-onset disease, especially among Hispanic/Latino patients. Early-onset colorectal tumors seemed to have more unique pathway activation than late-onset tumors in epithelial, immune, and stromal niches, as well as altered activation patterns. In the tumor microbiome of early-onset tumors, researchers found greater enrichment of bacterial species associated with cancer promotion.

Ancestry-related differences were also found in treatment responses, with distinct overall survival patterns for early- vs late-onset disease across chemotherapy, targeted therapy, and immunotherapy.

“These findings highlight the importance of incorporating ancestry and social determinants of health into biomarker discovery and treatment optimization, supporting more accurate implementation of precision medicine in young colorectal cancer populations with disproportionate health burdens,” the study authors concluded.

Pathway Alterations in FOLFOX-Treated Populations

Velazquez Villarreal also investigated pathway alterations in patients with colorectal cancers who were treated with the FOLFOX (fluorouracil, leucovorin, and oxaliplatin) chemotherapy to explore potential predictive biomarkers for both response to treatment and survival outcomes (Abstract 236). In a cohort of 2,515 patients with colorectal cancer from public datasets, the researchers analyzed somatic mutation and clinic data, focusing on alterations in the Wnt, TGF-beta, and PI3K pathways. The researchers used AI platforms to automate the construction of the cohort; stratification by age at onset, ancestry, and FOLFOX treatment status; and outcomes analyses.

The platforms were all adapted from the Artificial Intelligence Agent for High-Optimization and Precision Medicine (AI-HOPE) framework. The system, which was created in Velazquez Villarreal’s lab, uses an AI agent powered by large language models that was designed to advance precision medicine research by simplifying translational research.

Alterations in the Wnt pathway were frequently observed, with APC alterations being the most common. There appeared to be a significant difference in noncanonical CTNNB1 (5.5% vs 17.3%; P = .04) and RNF43 (5.5% vs 19.2%; P = .02) alteration frequencies for Hispanic/Latino patients with early-onset colorectal cancer who were vs were not treated with FOLFOX. Non-Hispanic White patients with late-onset colorectal cancer who were treated with FOLFOX showed reduced frequencies for AXIN1, AXIN2, RNF43, and TCF7L2 alterations. Wnt pathway alterations in non-Hispanic White patients were found to be associated with more favorable survival outcomes.

About 28% to 39% of Hispanic/Latino patients and 23% to 31% of non-Hispanic White patients had alterations in the TGF-beta pathway; alterations in SMAD4 were the most common, though these were more typically found among non-Hispanic White patients with early-onset disease than Hispanic/Latino patients (13.9% vs 2.7%). BMPR1A alterations seemed to be more common among Hispanic/Latino vs non-Hispanic White patients with early-onset disease who were treated with FOLFOX (5.5% vs 1.1%; P = .027). Alterations in SMAD2 and TGFBR2 were more frequently observed among untreated non-Hispanic White patients with late-onset disease. Overall, TGF-beta alterations were found to be associated with significantly worse overall survival in Hispanic/Latino patients with early-onset disease who were treated with FOLFOX (P = .029). As such, the researchers believe that alterations in the TGF-beta pathway could serve as ancestry- and treatment-specific biomarkers of poor prognosis in Hispanic/Latino patients with early-onset colorectal cancer who are treated with FOLFOX.

Alterations in the PI3K pathway appeared to vary by both age and ancestry. Among high-risk patients, the researchers reported that mutation-specific patterns emerged, suggesting INPP4B and RPTOR as potential biomarker candidates, although they were not statistically enriched. In non-Hispanic White patients with early-onset disease, PI3K alterations seemed to be linked to poor survival.

“Across all analyses, AI-HOPE automated the construction of biologically relevant subgroups, generated survival curves, computed mutation frequencies, and compared selected vs unselected cohorts with high precision and speed. These results demonstrate the platform’s value in identifying clinically actionable genomic signals, supporting ancestry-informed cancer research, and accelerating biomarker discovery in high-risk colorectal cancer populations,” Velazquez Villarreal and colleagues wrote in their poster.

Wnt Pathway Alterations in Bevacizumab-Treated Disease

In patients with colorectal cancer who were treated with the VEGF inhibitor bevacizumab, Velazquez Villarreal and colleagues looked specifically at Wnt alterations for possible prognostic biomarkers (Abstract 235). They retrospectively analyzed a cohort of 2,717 patients with colorectal cancer who had available genomic, clinical, and treatment annotations in public datasets. They then queried a Wnt-focused version of AI-HOPE to simplify multiparameter clinical questions.

According to the researchers, in Hispanic/Latino patients with early-onset colorectal cancer, those who were vs were not treated with bevacizumab had lower frequencies of RNF43 alterations (0% vs 14.3%; P = .03); this pattern was also seen in Hispanic/Latino patients with late-onset disease (2.3% vs 18.0%; P = .008). In non-Hispanic White patients with early-onset disease, exposure to bevacizumab appeared to lead to reduced frequencies of alterations in RNF43, TCF7L2, AXIN1, and AXIN2, whereas those with late-onset disease showed lower rates of AMER1, AXIN1, AXIN2, RNF43, and TCF7L2 alterations. Among patients who did not receive bevacizumab therapy, the researchers noted that earlier-onset groups tended to show higher rates of alterations in APC and TCF7L2 but lower frequencies of AXIN1, AXIN2, and RNF43 alterations.

Alterations in the Wnt pathway were found to be associated with improved overall survival in both Hispanic/Latino patients with early-onset colorectal cancer (P = .014) and non-Hispanic White patients with late-onset disease (P = .0052). In contrast, they appeared to predict worse overall survival outcomes in non-Hispanic White patients with early-onset disease (P = .023).

“Associations between bevacizumab and reduced Wnt mutation frequencies suggest potential interaction between antiangiogenic therapy and Wnt signaling,” the study authors concluded.

Conclusion

Collectively, the three analyses illustrate how biomarker relationships in colorectal cancer vary by age at diagnosis, ancestry, and treatment context, shining a light on the limitations of one-size-fits-all approaches to precision oncology. The studies reveal potential pathways and molecular features—using AI systems—that could inform risk stratification and treatment decision-making going forward for patients with colorectal cancer, particularly in those with early-onset disease and historically underserved populations.

Further prospective validation is needed to confirm the findings of these exploratory biomarker analyses, according to the researchers.

DISCLOSURE: Dr. Velazquez Villarreal reported no conflicts of interest.

ASCO AI in Oncology is published by Conexiant under a license arrangement with the American Society of Clinical Oncology, Inc. (ASCO^®). The ideas and opinions expressed in ASCO AI in Oncology do not necessarily reflect those of Conexiant or ASCO. For more information, see Policies.