Authors
Yasmein Yaser Salem, Joud Hourani, Ranim Al Rfai, Sara Abdelgawwad, Hamzah Dabool, Tasneem Zarzour, Rana Sammour
Published in
Drug development and industrial pharmacy. Pages 1-19. Nov 17, 2026. Epub Nov 17, 2026.
Abstract
To develop and experimentally validate an explainable artificial intelligence (AI) framework that extracts formulation rules for niosomal drug delivery systems from published data and delivers them through a decision-support platform.
Niosome performance depends on many interacting formulation and process variables, and optimization still relies largely on trial-and-error experimentation. Converting existing formulation knowledge into interpretable, experimentally verified rules could make early-stage development faster and less resource-intensive.
A database of 33 optimized niosomal formulations (approximately 30 drugs, BCS classes I-V) was compiled from peer-reviewed studies. Correlation analysis, XGBoost feature importance, and SHAP were used to relate formulation variables to particle size, entrapment efficiency (EE%), drug release, and release time. The extracted trends were tested experimentally using 17 diclofenac sodium-loaded niosomal formulations prepared by thin-film hydration. The dataset was expanded to 1000 samples using SMOTE to train XGBoost predictive models, which were evaluated on the original formulations and integrated with a Qwen-based large language model.
Phosphate-buffered saline as the hydration medium, Span 60, and chloroform volume were the most influential variables. Laboratory results agreed with the AI-predicted direction of effect for all tested variables: cholesterol, Span 60, chloroform volume, Tween 80, and hydration medium. The predictive models achieved R2 values of 0.97-0.99 on the original formulations, and the resulting Niosome AI Expert platform showed general agreement between predicted and reported outcomes.
Combining machine learning, explainable AI, and experimental verification can convert existing formulation knowledge into a practical decision-support tool that reduces reliance on trial-and-error niosome development.
PMID:
43043099
Bibliographic data and abstract were imported from PubMed on 17 Nov 2026.
Read full publication at:
Please sign in
to see all details.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 4
- Comments 0