Machine Learning-Based Exploration Of Phytochemical Pancreatic Lipase Inhibitors For Obesity Management

Abstract

Context: Obesity remains a critical global health issue, with pancreatic lipase inhibition emerging as a strategic target for therapeutic intervention. Phytochemicals have shown potential in modulating obesity-related pathways, particularly by inhibiting pancreatic lipase, a key enzyme in dietary fat metabolism.

Objectives: The current study aims to leverage machine learning (ML) techniques to study the effective phytochemicals thatinhibit pancreatic lipase and to model and predict IC50 values based on experimental parameters and to understand the influence of key descriptors like phytochemical concentration, exposure time, and medium.

Methodology: A dataset of 180 data points was compiled from 55 research articles (1988–2025) that include in vitro pancreatic lipase inhibition assay for various heterogenous phytochemicals. ML regression models—including Lasso, Random Forest, Gradient Boosting, SVR, and XGBoost—were trained on the dataset. Feature engineering included normalization, label encoding, and correlation-based feature selection.

Results: Lasso Regression achieved the highest R² score, indicating superior predictive performance. Key predictors influencing IC50 values included phytochemical concentration and assay time. Residual analysis confirmed minimal bias and strong model generalization. DMSO as a solvent and substrates like p-nitrophenyl butyrate significantly influenced outcomes.

Conclusion: In the present investigation, among the eleven predictive models evaluated, Lasso Regression emerged as the most efficacious in modeling pancreatic lipase activity based on the provided molecular descriptors. This underscores the potential of data-centric methodologies to streamline discovery pipelines in obesity therapeutics.