A Machine Learning Approach for Predicting Caco-2 Cell Permeability in Natural Products from the Biodiversity in...

Area of Science:

• Pharmacology and Drug Discovery
• Computational Chemistry
• Natural Products Research

Background:

• Peru possesses rich biodiversity and traditional knowledge of medicinal plants.
• Limited data exists on the intestinal absorption and permeability of these natural products.
• Caco-2 cell line assays are crucial for in vitro permeability assessment.

Purpose of the Study:

• To develop a quantitative structure-property relationship (QSPR) model for predicting Caco-2 cell permeability of Peruvian natural products.
• To utilize machine learning algorithms for accurate prediction of apparent permeability (log Papp).
• To assess the intestinal absorption potential of natural products from Peru.

Main Methods:

• Compiled a dataset of 1817 compounds with experimental log Papp values and molecular descriptors.
• Constructed and compared six QSPR models: MLR, PLS, SVM, RF, GBM, and a combined SVM-RF-GBM model.
• Evaluated model performance using RMSE and R² metrics on a testing set.

Main Results:

• The combined SVM-RF-GBM model achieved the best performance (RMSE = 0.38, R² = 0.76).
• The model predicted high intestinal absorption potential for 68.9% of 502 evaluated natural products.
• Natural products were categorized by metabolic pathways and assessed for drug-likeness.

Conclusions:

• The developed QSPR model effectively predicts intestinal absorption of Peruvian natural products.
• Findings facilitate the identification of promising natural products for drug development.
• This research supports pharmaceutical discovery efforts by highlighting compounds with favorable absorption characteristics.