|
|
 |
| |
|
|
|
Machine Learning-Based Prediction of Particle Size in PLGA Nanoparticle Formulations |
|
|
|
PP: 1345-1354 |
|
|
doi:10.18576/amis/200517
|
|
|
|
Author(s) |
|
|
|
Hamza Abu Owida,
Areen Arabiat,
Jaber A. Alwidian,
Hamza A. Mashagba,
Asokan Vasudevan,
Azlan B. Abd Aziz,
Mohammad Faleh Ahmmad Hunitie,
Suleiman Ibrahim Mohammad,
|
|
|
|
Abstract |
|
|
| Poly lactic-co-glycolic acid (PLGA) nanoparticles’ particle size is a crucial quality parameter, but precise pre-experimental prediction is still challenging since it depends on coupled polymer, drug, and formulation variables across heterogeneous investigations. Hydrodynamic particle size was modeled in this study using a carefully selected literature-derived dataset that included 430 formulations from 59 literature references and 15 predictors. Nevertheless, the suggested model shows very little cross-study generality. Drug hydrogen-bond acceptor count (mol Hacceptors), LA/GA ratio, surfactant concentration, drug/polymer ratio, drug lipophilicity, and polymer molecular weight were consistently found to be the main predictors of particle size by interpretable machine learning (ML) analyses, with distinct non-linear and interaction-dependent behavior. A grouped external hold-out split at the literature-reference level and random 5-fold cross-validation were used to assess the performance of the proposed model. While performance declined on the external reference-level test set (RMSE = 61.56 nm, MAE = 49.63 nm, R2 = 0.18), random 5-fold cross-validation demonstrated good apparent in-distribution performance (RMSE = 49.75 nm, MAE = 26.63 nm, R2 = 0.76). These results demonstrate that study-aware external validation is necessary prior to widespread predictive implementation, but they also support the use of interpretable machine learning for formulation screening and hypothesis creation. |
|
|
|
|
|
|
|
