AI-Driven Formulation Development: Integrating Machine Learning, Predictive Modeling, and Quality-by-Design for Next-Generation Pharmaceutical Products

Pharmaceutical formulation development has historically relied on empirical, trial-and-error experimentation supported by statistically designed experiments (DoE), a paradigm that is time-consuming, resource-intensive, and increasingly incompatible with the complexity of modern drug candidates, including poorly soluble small molecules, biologics, nucleic-acid therapeutics, and nanoparticulate delivery systems. Over the past decade, artificial intelligence (AI) and machine learning (ML) have emerged as complementary tools capable of learning non-linear structure-property and process-product relationships from historical and experimentally generated data, thereby narrowing the design space that must be explored empirically. In parallel, the Quality-by-Design (QbD) framework codified in International Council for Harmonisation (ICH) guidelines Q8 through Q14 has shifted pharmaceutical development and manufacturing toward a proactive, risk-based, and scientifically justified approach to quality assurance. This review synthesises the converging literature on AI/ML-directed formulation science and QbD, examining how predictive modelling, Bayesian optimisation, deep learning, and related computational techniques are being integrated into pre-formulation screening, excipient selection, process parameter optimisation, and real-time release testing across oral solid dosage forms, modified-release systems, nanoparticulate and lipid-based carriers, biologics, and lyophilised products. The review further evaluates predictive techniques for dissolution behaviour, physiologically based pharmacokinetics, and stability/shelf-life estimation, and considers how digital twins, process analytical technology (PAT), and self-driving laboratories are beginning to operationalise closed-loop, autonomous formulation development. Regulatory perspectives are examined through the U.S. Food and Drug Administration's 2023-2025 discussion papers and draft credibility-assessment guidance, alongside the broader Good Machine Learning Practice discourse. Persistent challenges are critically assessed, including data scarcity and heterogeneity, the opacity of “black-box” models, validation and reproducibility burdens, and workforce and infrastructural gaps, particularly in small and mid-sized manufacturers. The review concludes that AI/ML and QbD are not competing paradigms but increasingly interdependent ones: ML supplies the predictive and optimisation engine, while QbD supplies the risk-based scientific and regulatory scaffolding within which that engine can be trusted. Realising the full promise of AI-driven formulation development will depend on interdisciplinary collaboration, transparent and explainable modelling practices, harmonised regulatory frameworks, and sustained investment in shared, high-quality pharmaceutical datasets.