Machine learning-based analysis on pharmaceutical compounds interaction with polymer to estimate drug solubility in formulations.

This study introduces a sophisticated predictive framework for determining drug solubility and activity values in formulations via machine learning. The framework utilizes a comprehensive dataset consisting of more than 12,000 data rows and 24 input features containing a wide range of parameters to estimate drug solubility in formulation. The primary goal is to improve the accuracy of predictions by using ensemble learning techniques. Three base models were evaluated including: Decision Tree (DT), K-Nearest Neighbors (KNN), and Multilayer Perceptron (MLP), which are subsequently improved with the AdaBoost ensemble method. To further optimize performance, Recursive Feature Elimination (RFE) is employed for feature selection with the number of features treated as a hyperparameter. Hyperparameter tuning is rigorously conducted utilizing the Harmony Search (HS) algorithm. For drug solubility prediction, the ADA-DT model demonstrates superior performance, achieving an R² score of 0.9738 on the test set, with a Mean Squared Error (MSE) of 5.4270E-04 and a Mean Absolute Error (MAE) of 2.10921E-02. For gamma prediction, the ADA-KNN model outperforms other models, with an R² value of 0.9545 on the test set, an MSE of 4.5908E-03, and a MAE of 1.42730E-02. The results show that ensemble learning with advanced feature selection and hyperparameter optimization can accurately predict complex biochemical properties.