Application of Artificial Intelligence in Predicting the Solubility of Pharmaceutical Structures Using Electronic Properties an‎d Vibrational Frequencies

5/28/2025 12:00:00 AM

In the present study, a hybrid framework based on density functional theory (DFT) computations an‎d machine learning algorithms was developed to predict the solubility of pharmaceutical compounds. In the first step, electronic properties were extracted using Gaussian software an‎d the DFT method, an‎d vibrational frequencies along with influential parameters related to these frequencies—such as zero-point energy—were considered for solubility analysis. Moreover, the main frequency present in each pharmaceutical structure, particularly those containing functional groups like –OH, –NH, an‎d structures with strong bonding groups capable of significant interactions with solvents, were independently utilized in the eva‎luation. In the next phase, by applying feature engineering methods such as data normalization, the dataset was optimized for training machine learning models. Several algorithms including multiple linear regression, decision tree, support vector machine, gradient boosting, an‎d XGBoost were trained an‎d eva‎luated using cross-validation. The results show that XGBoost an‎d ran‎dom forest models outperform others by achieving the highest coefficient of determination an‎d the lowest mean squared error. Feature importance analysis revealed that electronic properties an‎d the intensity of vibrational frequencies significantly influence solubility prediction. The findings of this study indicate that vibrational spectroscopy an‎d computational eva‎luation of electronic properties can play an effective role in estimating drug solubility.

حلاليت , ماشين لرنينگ

Solubility , Machine learning

Fatemeh Sobhani

Dr. Majid Hashemianzadeh