(T1130-01-02) Application of Knowledge Based Expert System to Predict API-Excipient Incompatibilities

Purpose: API-excipient incompatibilities can complicate a product life cycle by causing unwanted interactions in the final dosage form thus losing therapeutic value of the product. It is one of the critical challenges that scientist faces during formulation design. Expert System is a knowledge base app to predict API excipient incompatibilities. Expert system app will ease the formulation design phase by predicting obvious incompatible excipients at the early formulation development stage.

Methods: A knowledge-based app can be defined as an application developed based on knowledge found in literature. A database had been developed consisting of all the expected reactions (such as Acid base reaction, hydrolysis reaction etc.) of any two functional groups present in the API, excipient and/or impurities present in the excipient. A total of 16 reactions that has been reported in the past is enlisted in the database. This database is called rule database as it sets up the rule. The expert system application derives this reaction rule/knowledge and uses it to predict incompatibilities between API and excipients based on their chemical structures. The app consists of a graphical interface where a user can input a drug/API structure. On submitting the API information, the app will go through the rule database to see if the functional group present in the API has any rule match. If there is a positive match, the search then goes onto an excipient database where 28 commonly used excipients information has been stored along with their impurities details. After the positive match through the rule database between an API with an excipient/impurity the calculation further happens based on the molecular weight and weight percentage of the participant molecules to understand the extent of the reaction. This kind of semiquantitative prediction alleviate the burden of unwanted obvious reactions that has been reported before and can save significant time in early formulation development. This application is inspired from PharmDE platform [1], which predicts API-excipient incompatibilities as is. However, Expert system has been developed beyond this capacity to predict weighted risk between API-excipient based on not only their structure but also molecular weight and their proportion present in a formulation. Based on this calculation, the API-excipient interaction provides more appropriate risk approximation. Moreover, Expert system is also able to predict nitrosation risk of an API based on their structure [2]. Pycharm community edition 2021.3.2 has been used as a platform to develop the app using flask and python 3.8 and the reactions, as well as excipient database has been stored in MS Excel. The application consists of web apps where the user can provide API SMILES structure and click the submit button. Upon submission, the app finds out the functional group present in the API structure via a substructure match using RDkit and fetches the plausible interactions caused by the functional group of the API with the functional group present in the excipients/impurities.

Results: Two types of validation run have been performed to validate the Expert system platform. i. 4 known API candidates (Moexipril, Acyclovir, Aspirin and Gabapentin) have been selected and has been run as case studies in both PharmDE and Expert system platform and the results were compared (Table 2). ii. Experimental data found on Wang et. Al has been run in expert system and the results were compared qualitatively.

Conclusion: Validation runs were carried out between PharmDE and Expert system to establish a baseline for API-excipient incompatibilities. A total of 4 known API-excipient compatibility study was run and the similarity has been confirmed for all of them. Based on this,  the platform has been validated and currently is being used at preformulation stage in the company. This research lays the foundation for improving the dosage development at preformulation stage and gather more information on a new molecule and avoid potential future delays, which will in turn improve patient compliance, lower drug price, and fulfill public health mission.

References: 1. Wang, N., Sun, H., Dong, J., & Ouyang, D. (2021). PharmDE: A new expert system for drug-excipient compatibility evaluation. International Journal of Pharmaceutics, 607, 120962.
2. Schlingemann, J., Burns, M. J., Ponting, D. J., Avila, C. M., Romero, N. E., Jaywant, M. A., ... & Wilk, A. (2023). The landscape of potential small and drug substance related nitrosamines in pharmaceuticals. Journal of Pharmaceutical Sciences, 112(5), 1287-1304.

Acknowledgements: We are thankful to our summer intern Dushyanth Gopichand. We are also thankful to PharmDE developer for developing the platform and providing public access.


Fig 1: Expert_system_workflow


Table 1: Rule table example


Table 2: Validation run result