(W0930-06-37) Artificial Intelligence to Predict Drug Release from Sustained-Release Coated Tablets

Purpose: Artificial intelligence (AI) is transforming the pharmaceutical sciences by accelerating drug discovery and product development by enabling performance predictions with minimal experimentation. The FDA recently issued a guidance document on the use of AI in pharmaceutical research (1). This study aimed to develop AI predictive models, specifically Artificial Neural Networks (ANN) and Random Forest (RF), to predict drug release from sustained-release coated tablets. ANN utilizes forward and backward propagation with weighted variables to model complex patterns by mimicking brain-like neuron layers. In contrast, RF uses an ensemble of decision trees to make predictions through majority rule or averaging. Eudragit^® RS and Eudragit^® RL polymethacrylates are insoluble polymers commonly used to coat tablets and slow drug release. These polymers differ by the number of quaternary ammonium functional groups, with the RL polymer having a greater number of these functional groups thus being more permeable than the RS polymer. Combinations of the polymers are often used to achieve different permeabilities and thus different drug release rates. Drug release is also affected by the thickness of the coating applied. The objective of this study was to develop robust machine learning models to predict drug release from sustained-release coated tablets based on experimental data. Variables investigated included coating levels and ratios of the two Eudragit polymers.

Methods: Tablets containing acetaminophen (APAP) as the model drug were prepared using a single station tablet press. Tablets were then coated in Vector LDCS perforated coating pan apparatus with combinations of Eudragit^® RS and RL ranging from 5:95 to 80:20. Triethyl citrate at a concentration of 20% (based on dry polymer weight) was added to the Eudragit suspensions and water was added to reduce the solids content to 20% (w/w). Spray rate was approximately 4.5g/min and the inlet temperature was set at 48°C. Coating levels were varied from 3 to 10% weight gain (based on dry polymer weight). Coated tablets were cured at 40°C for 2 hours after coating. A USP dissolution apparatus II was used to determine drug release over time in pH 6.8 phosphate buffer. Drug concentrations were quantified using HPLC (2). Drug release data were used to build machine learning models. Drug release predictions were then compared to actual data.

Results: Drug release data from tablets coated at the 3% weight gain level exhibited considerable variability, most likely due to non-uniform coating application. Consequently, this dataset was excluded from model development. Drug release predictions generated by both ANN and RF models for the 10:90 RS:RL ratio at a coating level of 8.5% were similar and in agreement with experimental data (Fig. 1). The two models were also used to predict drug release for a 40:60 RS:RL polymer ratio at 5% and 10% coating levels (Fig. 2). The drug release profiles were similar at the 5% coating level, with ANN predicting slightly higher release after 6 hours. In addition, the predictions at the 10% coating levels were nearly identical over the 8 hour time period.

Conclusion: While ANN and RF differ in the learning approach, both models demonstrated strong predictive performance for drug release from sustained-release film-coated tablets. These findings suggest that AI models can serve as reliable tools for predicting drug release from coated tablets, potentially reducing the need for extensive empirical testing during formulation development.

References: 1. U.S. Food and Drug Administration (2025). Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products: Guidance for Industry and Other Interested Parties (Draft Guidance).
2. Felton, L. A., Binzet, G., Wiley, C., McChesney, D., McConville, J., Çelik, M., & Muttil, P. (2024). Spray drying Eudragit® E-PO with acetaminophen using 2- and 3-fluid nozzles for taste masking. International Journal of Pharmaceutics, 658, 124191.

Acknowledgements: This work was supported in part by an NIH grant (R21HD095100) and the UNM Comprehensive Cancer Center Support Grant NCI P30CA118100. We would like to thank Evonik for donating the polymers used in this study.


Figure 1. Comparison of ANN and RF model predictions with experimental drug release data for 10:90 RS:RL polymer ratio at 8.5% (w/w) coating level.


Figure 2. Comparison of drug release predictions from ANN and RF models at 40:60 RS:RL polymer ratio. (A) 5 % (w/w) coating level; (B) 10% (w/w) coating level.