Computational insights into drug hygroscopicity by coupling machine learning and molecular simulation

Area of Science:

• Pharmaceutical Sciences
• Computational Chemistry
• Drug Development

Background:

• Hygroscopicity is a critical material attribute for active pharmaceutical ingredients (APIs) that impacts manufacturability, stability, and efficacy.
• Traditional experimental hygroscopicity measurements are time-consuming, hindering efficient preformulation screening.
• High-throughput computational methods are needed for rapid identification of hygroscopic compounds in drug discovery.

Purpose of the Study:

• To develop and validate an integrated computational strategy for predicting drug hygroscopicity and elucidating its underlying mechanisms.
• To compare the performance of various machine learning algorithms for predicting moisture-induced weight change in drugs.
• To identify key molecular descriptors influencing hygroscopicity and understand water-molecule interactions through simulations.

Main Methods:

• Curated a dataset of dynamic vapor sorption (DVS) curves for 607 drugs.
• Evaluated 8 machine learning algorithms, including Tabular Prior-data Fitted Networks (TabPFN), for regression and classification tasks.
• Employed SHapley Additive exPlanations (SHAP) for feature importance analysis and utilized molecular dynamics and quantum chemical simulations for mechanistic insights.

Main Results:

• TabPFN demonstrated superior performance in predicting moisture-induced weight change (R² of 0.701 ± 0.075) and classifying hygroscopicity.
• SHAP analysis identified molecular surface area, polarity, and electrostatic descriptors as key predictors of hygroscopicity.
• Molecular simulations confirmed that polar functional groups, hydrogen bonding, and surface conformation dictate water interactions.

Conclusions:

• The integrated computational approach effectively predicts drug hygroscopicity and provides mechanistic understanding.
• This AI-driven strategy combined with physics-based simulations offers a powerful tool for preformulation developability screening.
• The proposed method has the potential to significantly reduce research and development costs and improve drug development efficiency.