AI-Compass: Development of an AI-Driven Predictive Tool for Drug-Excipient Compatibility Assessment with FTIR and DSC supported Experimental Validation for Rational Preformulation Design

Background

Selecting the right excipients is essential for developing stable and effective pharmaceutical formulations. However, identifying potential drug-excipient incompatibilities using conventional methods can be time-consuming, labour-intensive, and costly. This creates a need for faster and more efficient approaches, especially during the early stages of formulation development. The present study focuses on developing an artificial intelligence-based tool, DE-Interact, designed to quickly predict drug-excipient compatibility and support rational formulation decisions.

Objective

The aim of this study was to develop and validate an AI-based prediction tool to determine drug-excipient compatibility at an early stage in development (DE-Interact). This tool has been developed based on molecular fingerprint data from PubChem and artificial neural network models. The interactions were then predicted and confirmed experimentally by Differential Scanning Calorimetry (DSC) and Fourier Transform Infrared Spectroscopy (FTIR) analysis, and the ultimate aim was to develop the rational data-driven method for preformulation decision making that is fast.

Materials and Methods

DE-Interact was built using molecular fingerprints obtained from PubChem and trained through an artificial neural network on a curated dataset of drug-excipient combinations. Several combinations such as Ketoprofen + Magnesium Stearate, Ketoprofen + Magnesium oxide, Ketoprofen + PEG 400, Enalapril + Aerosil, and Aceclofenac + Magnesium stearate were selected as case studies. The combinations were evaluated using standard performance metrics and further validated experimentally using techniques such as DSC, FTIR.

Results

The model showed strong and accurate predictive capability of identifying incompatible combinations. These predictions were supported by experimental findings, which revealed changes such as altered thermal behaviour, reduction or disappearance of characteristic functional group peaks, and signs of degradation. Importantly, the study also captured time-dependent incompatibilities that may not be immediately visible but can affect long-term stability.

Conclusions

DE-Interact offers a practical and efficient solution for pre-formulation compatibility screening. By combining artificial Intelligence with the reliability of experimental validation, researchers can reduce development time, optimize resource utilization, and improve decision-making in formulation design. This integrated approach represents a meaningful step toward more streamlined and data-driven pharmaceutical development.

Graphical Abstract