Advancing Direct Tablet Compression with AI: A multi-task framework for quality control, batch acceptance, and causal analysis

Abstract

Pharmaceutical manufacturing has surged in drug development with the rise of Pharma 4.0, leveraging artificial intelligence (AI) to improve efficiency, optimize resource use, and reduce production times. Direct Tablet Compression (DTC), a key manufacturing technique, depends on the physicochemical properties of active pharmaceutical ingredients (API), excipients, and process parameters. This paper presents a novel multi-task framework combining regression, classification, and text generation to predict tablet properties (friability, hardness, disintegration time, and water absorption ratio), determine batch acceptance, and provide insights for optimizing interactions to improve tablet quality. The framework not only enables real-time monitoring, quality control and regulatory compliance, but also helps to understand the reasons why tablets in the manufacturing batch do not meet quality requirements. Using statistical methods, Neural Networks (NN), Natural Language Processing (NLP), and generative AI (GenAI), it outperforms state-of-the-art methods, achieving 91.8% $R^2$ and 95.5% accuracy for regression and classification, respectively, as demonstrated using the Harvard Dataverse V1 dataset of Fast Disintegrating Tablets (FDTs) non placebo.