ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING FOR PREDICTING PHYTOCHEMICAL-INDUCED APOPTOSIS IN BREAST CANCER CELL LINES: A MOLECULAR APPROACH TO CANCER PREVENTION

Abstract

Background: Phytochemicals are increasingly being studied for their potential role in breast cancer research because many of these naturally derived compounds can influence cancer-related cellular processes. However, testing large numbers of phytochemicals experimentally is expensive, time-consuming, and difficult to scale. This study therefore explored whether machine learning could be used to predict phytochemical-induced apoptosis and help identify compounds that may deserve further experimental investigation. Methods: A dataset containing 1,347 experimental observations collected from 54 peer-reviewed studies was prepared using information from breast cancer cell-line experiments. The variables included phytochemical type, concentration, treatment duration, breast cancer cell line, combination treatment, cell viability, and apoptosis. Ten regression models were evaluated, including Linear Regression, Support Vector Regression, K-Nearest Neighbors, Decision Tree, Gradient Boosting, LightGBM, CatBoost, Random Forest, XGBoost, and Extra Trees. Association Rule Mining was also applied to identify recurring relationships among the experimental variables, while SHAP analysis was used to explain how individual features contributed to model predictions. Results: The findings showed that tree-based ensemble methods performed better than the simpler statistical and conventional machine-learning approaches. Extra Trees produced the highest reported training performance with an R² of 0.9853, followed closely by XGBoost (R² = 0.9852) and Random Forest (R² = 0.9839). Among the evaluated phytochemicals, curcumin showed the highest average apoptosis value at 52.87%, followed by resveratrol at 46.13%. Association Rule Mining revealed recurring patterns linking higher phytochemical concentrations with increased apoptosis and reduced cell viability. SHAP analysis further showed that phytochemical type had the strongest influence on apoptosis prediction, followed by breast cancer cell line, cell viability, and concentration. Conclusion: The study demonstrates that combining machine learning with explainable AI and association rule mining can provide a practical approach for analysing phytochemical responses in breast cancer cell-line experiments. The framework may help researchers narrow down promising phytochemicals before conducting further laboratory testing. Nevertheless, the findings should be validated using independent datasets and additional experimental studies before the approach is considered for translational or clinical use.