Abstract B005: Machine learning-enhanced association analysis of ancestry-informative marker SNPs identifies predictors of head and neck cancer status in an underserved population

Abstract Background: Head and neck squamous cell carcinoma (HNSCC) exhibits persistent disparities in incidence, treatment, and survival among underserved populations. Although tobacco use, alcohol consumption, and human papillomavirus infection are established risk factors, inherited genetic variation also contributes to disease susceptibility. We investigated ancestry-informative marker (

Aim

single nucleotide polymorphisms (SNPs) and supervised machine learning to identify genetic and clinical predictors of HNSCC. Methods: A retrospective case-control study included 905 participants (209 HNSCC cases, 696 controls). Clinical variables included age, sex, marital status, smoking status, alcohol use, and insurance status.

Aim

SNPs were coded using an additive model according to African ancestral allele dosage (0 = no African ancestral allele, 1 = one African ancestral allele, 2 = two African ancestral alleles). Quality control included filtering for missingness, minor allele frequency, Hardy-Weinberg equilibrium, and genotype imputation. Multivariable logistic regression adjusted for clinical covariates identified associated SNPs. Significant SNPs and clinical variables were incorporated into Logistic Regression, Random Forest, Gradient Boosting, XGBoost, and Support Vector Machine models using an 80:20 train-test split with five-fold cross-validation. Performance was assessed by accuracy, sensitivity, specificity, precision, F1-score, and ROC-AUC. Results: Logistic regression identified multiple

Aim

SNPs associated with HNSCC after covariate adjustment. Random Forest demonstrated the best performance, correctly classifying 214 of 227 test samples with 94.3% accuracy, 80.8% sensitivity, 98.3% specificity, 93.3% precision, and an F1-score of 86.7%. Feature importance analysis showed that marital status, smoking status, alcohol use, insurance status, and multiple

Aim

SNPs—including rs4149436, rs188574574, rs4236541, rs1011024, rs1058491, rs2740566, rs5945361, rs5987175, rs10953286, and rs888096—were among the strongest predictors, demonstrating that ancestry-associated genetic variation complements established clinical risk factors. Conclusions: Integrating

Aim

SNPs with supervised machine learning improved prediction of HNSCC while identifying both genetic and social determinants associated with disease. Coding SNPs by African ancestral allele dosage enabled evaluation of ancestry-associated genetic variation within predictive models. These findings demonstrate the potential of combining ancestry-informed genomics, clinical characteristics, and social determinants of health to improve risk stratification and advance equitable precision oncology for populations disproportionately affected by HNSCC. AI Disclosure Statement: ChatGPT (OpenAI) was used solely to assist with language editing and organization. All study design, data analysis, machine learning, interpretation, and scientific conclusions were performed, verified, and approved by the authors. Citation Format: Maxine Harlemon, Jeffery Liu, Zhongxuan He, Yan Zhou, Elizabeth Blackman, Denise Gibbs, Sharon Harrison, Karthik Devarajan, Camille Ragin. Machine learning-enhanced association analysis of ancestry-informative marker SNPs identifies predictors of head and neck cancer status in an underserved population [abstract]. In: Proceedings of the 19th AACR Conference on the Science of Cancer Health Disparities; 2026 Oct 18-21; Atlanta, GA. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2026;35(10 Suppl):Abstract nr B005.