(M1330-08-53) Machine Learning and Automation Accelerate Formulation Development by Enabling Fast High-Throughput Screening for Long-Acting Injectable Drug Formulations

Purpose: Machine learning (ML) predictive models and laboratory automation are advancing the development of pharmaceutical and non-pharmaceutical formulations by enhancing efficiency, accuracy, and innovation. ML algorithms analyze large datasets to identify patterns and predict outcomes, significantly speeding up the development process. This leads to faster identification of relevant parameter variables, optimization of formulations, and prediction of their behavior in various conditions. Automation complements this process by streamlining repetitive and complex tasks, such as high-throughput formulations building and testing. Automation can increase throughput and allow the fast generation of large datasets needed for ML model training. Together, ML and automation enable rapid prototyping, real-time data analysis, and adaptive learning, ultimately accelerating the development cycle, reducing costs, and improving the precision and effectiveness of formulations. This is particularly useful in the development of sustained-release PLGA microsphere drug formulations where particular formulation parameters result in variable drug loading and release kinetics.

Methods: We have developed a platform that allows rapid PLGA microparticle formulation building, which allows high-throughput screening of the formulation parameters (e.g., polymers, concentrations, additives, flow times). The platform is used to collect data about the formulations’ physicochemical characterization, including encapsulation efficiency, drug loading, particle size distribution, and in vitro release profiles using dissolution assays. The development is assisted by the use of ML predictive modeling that uses an iterative approach to allow faster and more accurate target performance optimization. The observed performance of the first set of formulations is used as the dataset for predictive model training. The model predictions are used to design the second set of formulations that also serve as the retraining dataset for the next iteration - until targeted PPM performance is achieved.

Results: This iterative approach has been used for PLGA microsphere development for drugs, including hydrophobic small molecules and biologics. We were able to optimize formulations to achieve targeted drug loading (i.e., over 20%) in particles and optimized particle characteristics to achieve specific release kinetics (i.e., one month).

Conclusion: We demonstrate the use of this process in the development of small drug and biologics formulations that resulted in 8-10x acceleration of the screening process over conventional development methods.

Process of formulation optimization development using iterative formulation building and ML model training

Formulation parameters used in the PLGA microsphere formulation optimization

Optimization of drug loading and release profile in PLGA microspheres with hydrophobic small molecule payload