Designing Optimum Drug Delivery Systems Using Machine Learning Approaches: a Prototype Study of Niosomes.

Area of Science:

• Pharmaceutical Sciences
• Computational Chemistry
• Drug Delivery Systems

Background:

• Niosome formulations are crucial for drug delivery.
• Optimizing niosome characteristics like particle size and drug entrapment is essential for efficacy.
• Traditional methods for formulation optimization can be time-consuming and resource-intensive.

Purpose of the Study:

• To develop and validate a machine learning model for optimizing niosome drug formulations.
• To identify key formulation parameters influencing niosome properties.
• To compare the predictive performance of artificial neural networks (ANN) against response surface methodology (RSM).

Main Methods:

• Systematic literature review following PRISMA guidelines to gather data on 114 niosome formulations.
• Utilized an artificial neural network (ANN) with a hyperbolic tangent sigmoid transfer function and Levenberg-Marquardt backpropagation for model training.
• Employed sensitivity analysis to determine critical formulation factors.
• Validated the ANN model through the preparation of Donepezil hydrochloride niosome batches using a 3x3 factorial design.

Main Results:

• The ANN model achieved high prediction accuracies: 93.76% for % drug entrapment and 91.79% for particle size.
• Sensitivity analysis identified drug/lipid ratio and cholesterol/surfactant ratio as the most significant factors.
• The validated model demonstrated over 97% prediction accuracy for experimental niosome batches.
• The ANN model outperformed RSM in predicting Donepezil niosome formulation parameters.

Conclusions:

• Machine learning, specifically ANN, offers a powerful and accurate approach for optimizing niosome drug formulations.
• Key formulation parameters like drug/lipid and cholesterol/surfactant ratios significantly impact niosome characteristics.
• The developed ANN model shows promise for the rational design of novel niosomal drug delivery systems, though further validation with diverse drugs is recommended.