A Generative\nAdversarial Network Approach to Predict\nNanoparticle Size in Microfluidics

To achieve precise control over the properties and performance\nof nanoparticles (NPs) in a microfluidic setting, a profound understanding\nof the influential parameters governing the NP size is crucial. This\nstudy specifically delves into poly(lactic-<i>co</i>-glycolic\nacid) (PLGA)-based NPs synthesized through microfluidics that have\nbeen extensively explored as drug delivery systems (DDS). A comprehensive\ndatabase, containing more than 11 hundred data points, is curated\nthrough an extensive literature review, identifying potential effective\nfeatures. Initially, we employed a tabular generative adversarial\nnetwork (TGAN) to enhance data sets, increasing the reliability of\nthe obtained results and elevating prediction accuracy. Subsequently,\nNP size prediction was performed using different machine learning\n(ML) techniques including decision tree (DT), random forest (RF),\ndeep neural networks (DNN), linear regression (LR), support vector\nregression (SVR), and gradient boosting (GB). Among these ensembles,\nDT emerges as the most accurate algorithm, yielding an average prediction\nerror of 8%. Further simulations underscore the pivotal role of the\nsynthesis method, poly(vinyl alcohol) (PVA) concentration, and lactide-to-glycolide\n(LA/GA) ratio of PLGA copolymers as the primary determinants influencing\nNP size.