Abstract 7449: Evaluating the effectiveness of machine learning models using polygenic risk scores and clinical factors for breast cancer risk stratification

Abstract Breast cancer is the most common cancer among women in the United States, aside from skin cancer. Early detection and treatment significantly reduce mortality rates, highlighting the importance of effective risk prediction methods. This study evaluates the predictive performance of multiple machine learning models that include polygenic risk scores (PRS), clinical and demographic factors. We compared three predictive models using 313 SNPs model from PGS catalog and incorporated age, body mass index (BMI), five genetic principal components (PC1-5), and breast density as predictor variables. Logistic Regression (LR), Multilayer Perceptron (MLP), and Probabilistic Graphical Model (PGM) were implemented to assess breast cancer risk. MLPs are a type of artificial neural network that consists of layers of interconnected "neurons" that work together to process and learn from data. Each layer processes the information and passes it to the next layer, allowing the model to learn complex patterns and detect non-linear relationships in the data, which makes it useful for predictions. The selected PGM implementation uses Bayesian statistics to determine feature importance and correlation. The model then uses this information to predict the probability an instance belongs to a group (e.g. case or control). Among 3, 000 African ancestry women from the Penn Medicine BioBank, model performance varied. Logistic Regression demonstrated the highest predictive accuracy at 0.69, followed by Probabilistic Graphical Model at 0.64 and Multilayer Perceptron at 0.63. Our study suggests limited interaction between PRS and clinical factors in breast cancer prediction. These findings aligns with prior research demonstrating the largely independent nature of polygenic risk scores from traditional clinical risk factors. These findings underscore the challenges in developing more sophisticated risk prediction models and highlight the need for alternative approaches to capture the complex genetic architecture of breast cancer risk Subsequent research will incorporate moderate and high penetrance genes to enhance the predictive power of our risk assessment models, potentially improving precision in breast cancer risk stratification. Citation Format: Lannawill Caruth, Roukiatou Sore, Sarah Ehsan, Anne-Marie McCarthy, Shefali Setia-Verma. Evaluating the effectiveness of machine learning models using polygenic risk scores and clinical factors for breast cancer risk stratification [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 7449.