Machine learning–driven prediction of in vitro dissolution profiles for glipizide solid dispersions
Abstract
In vitro dissolution profiling is central to developing oral dosage forms of poorly water-soluble drugs. Glipizide shows solubility-limited performance, so polymer-based solid dispersions (SDs) (polyvinylpyrrolidone (PVP) K30, PVP K90, polyethylene glycols (PEG) 6000) are used to enhance dissolution. Using percent-release profiles digitized from a published source, dissolution trajectories from 16 formulations—spanning multiple drug: polymer ratios plus reference tablets—were modeled from long-format time-series. Features comprised formulation metadata, engineered time descriptors, and optional kinetic-fit descriptors. Five approaches—bagged trees, least squares boosting (LSBoost), support vector regression (SVR), Gaussian process regression (GPR), and a long short-term memory (LSTM) sequence-to-sequence model—were compared using grouped 5-fold cross-validation by formulation to prevent time-point leakage, with performance assessed by root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). Bagged trees (RMSE 18.90%; R² 0.701) and the LSTM (RMSE 19.18%; R² 0.692) performed best and were statistically indistinguishable, while LSBoost was moderate and SVR and GPR underfit, producing near-constant predictions and negative R². Curve overlays showed the leading models preserved polymer-family separation and key dissolution-shape features, with occasional formulation-level outliers under strict holdout. Under formulation-level generalization, tree ensembles and sequence models predict glipizide SD dissolution with useful accuracy, supporting more efficient screening and rational optimization.
Keywords
Dissolution profile; Glipizide; Machine learning; Predictive modeling; Solid dispersion
Full Text:
PDFDOI: https://doi.org/10.11591/eei.v15i6.13199
Refbacks
- There are currently no refbacks.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
Bulletin of Electrical Engineering and Informatics (BEEI)
ISSN: 2089-3191
,
e-ISSN: 2302-9285
This journal is published by the
Institute of Advanced Engineering and Science (IAES)
.