Machine learning-driven insights into the mechanical performance of polymeric microneedle array patches

With growing interest in microneedle technology for enhancing transdermal drug delivery, the current body of literature provides a valuable data reservoir for understanding and predicting mechanical performance - a critical prerequisite for microneedle array patches (MAP) to effectively breach the skin barrier. The integration of artificial intelligence and machine learning (ML) has generated considerable interest in pharmaceutical research, offering the potential to streamline formulation development by reducing experimental workload and time. In this study, ML algorithms were applied to predict the mechanical properties of polymeric MAP using literature-mined data. A total of 313 formulations were extracted from 148 publications, with gradient boosting identified as the most robust model for predicting failure force and in vitro skin insertion, yielding errors of ±0.467 N and ± 7.781 %, respectively. Ex vivo skin penetration data were excluded owing to limited data availability. The models highlighted the predominant influence of microneedle dimensions, particularly inter-needle spacing, needle height and volume, along with polymer content and molecular weight in determining mechanical performance. Overall, this work demonstrates the utility of ML in developing predictive models for mechanical properties of polymeric MAP from existing literature datasets.