FormulationLAI: A physiology-based machine learning framework for accelerated development of long-acting injectable...

Long-acting injectables (LAIs) represent promising drug delivery platforms for chronic diseases management, but their clinical translation remains constrained by extremely long trial-and-error experiments (8-10 years), limited mechanism insights, and poor in vitro-in vivo correlation (IVIVC). How can we design LAI formulations more efficiently and predictably? Here, we propose a closed-loop computational framework integrating machine learning (ML), physiologically based pharmacokinetic (PBPK) modeling, and molecular dynamics (MD) simulations to accelerate LAI formulation design. Using the commercial poly (lactic-co-glycolic acid) (PLGA)-based injectable Zilretta® as a benchmark, PBPK/PD modeling was used to define target in vitro release profiles that maximize therapeutic exposure. Guided by ML predictions, an optimized in situ forming gel (ISFG) formulation was developed for intra-articular delivery of triamcinolone acetonide, achieving a 34 % increase in drug exposure compared to commercial suspension in rat pharmacokinetic studies. MD simulations elucidated key drug-polymer interactions enabling sustained release. The transformative computational framework could significantly reduce the formulation development time during the preclinical phase from the traditional 5-7 years to approximately 1-2 years. This framework redefines the formulation paradigm: from empirical "formulate-and-test" to predictive "compute-and-validate", and offers a scalable strategy for rational, efficient development of long-acting injectables.