Machine Learning for In Silico Predictive Modeling of Drug-Excipient Compatibility for Formulation Development

Review Article

Abstract

The selection of pharmaceutical excipients is the main determinant of the stability, safety, and therapeutic efficacy of drug products. Unfavorable drug–excipient interactions lead to chemical degradation or physical instability, compromising formulation integrity. Traditional empirical preformulation screening methods, though reliable, are constrained by high costs, low throughput, and prolonged timelines. The use of artificial intelligence and machine learning offers a powerful alternative, enabling rapid, data-driven compatibility predictions. This review discusses about the computational tools specifically DE-Interact, PharmDE, and ExPreSo highlighting their underlying systems, algorithms, and predictive capabilities. DE-Interact utilizes artificial neural networks trained on combined molecular fingerprints to deliver binary compatibility classifications for small molecules. PharmDE functions as a hybrid expert system, combining a curated database of reported incompatibilities with a rule-based inference engine to categorize formulations into high, medium, or low risk. For macromolecular systems, ExPreSo implements supervised ensemble machine learning utilizing protein sequence embeddings, structural features, and product attributes to predict excipient occurrence in stable formulations. While these tools significantly accelerate preformulation screening and minimize resource expenditure, hurdles remain regarding dataset size, representation of rarely used excipients, and regulatory integration. Improving model robustness through expanded, high-fidelity datasets and structural descriptors will be important to establishing fully automated, predictive pipelines in rational pharmaceutical formulation development