A water solubility prediction algorithm based on the StackBoost model.

Area of Science:

• Computational Chemistry
• Machine Learning
• Drug Discovery

Background:

• Aqueous solubility is a critical physicochemical property with broad applications.
• Experimental solubility determination is resource-intensive.
• Accurate solubility prediction is crucial for various scientific domains.

Purpose of the Study:

• To develop and evaluate a novel ensemble learning model, StackBoost, for predicting the aqueous solubility of organic compounds.
• To compare the performance of StackBoost against established ensemble methods.
• To assess the model's applicability in high-throughput screening and its generalization capabilities.

Main Methods:

• Development of the StackBoost model.
• Systematic comparison with Adaptive Boosting (AdaBoost), Gradient Boosted Regression Trees (GBRT), Light Gradient Boosting Machine (LGBM), Extreme Gradient Boosting (XGBoost), and Random Forest (RF).
• Validation using high-throughput screening on large datasets and transfer learning.

Main Results:

• StackBoost achieved a coefficient of determination (R²) of 0.90, RMSE of 0.29, and MAE of 0.22.
• StackBoost significantly outperformed all comparative ensemble models.
• High-throughput screening successfully identified compounds with high water solubility potential.
• The model demonstrated considerable transferability across datasets, indicating good generalization ability.

Conclusions:

• StackBoost is a highly effective model for predicting aqueous solubility.
• The model offers a resource-efficient alternative to experimental methods.
• StackBoost shows promise for large-scale screening and generalizable solubility prediction.