1. Data Mining and Machine Learning Approaches in Designing Optimum Drug Delivery Systems: A Prototype Study of Niosomes
- Author
-
Samar Salam Qawoogha, Nuruzzaman Faruqui, and Aliasgar Shahiwala
- Subjects
Computer science ,business.industry ,Drug delivery ,Artificial intelligence ,Niosome ,Machine learning ,computer.software_genre ,business ,computer - Abstract
This study aims to model the prediction of two clinically relevant properties of drug delivery systems specifically for niosomes, i.e., particle size and drug entrapment efficiency (EE%) using a combined approach of data mining, artificial neural networks (ANN), and design of experiments (DoE). Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) system was adopted to screen published literature on niosomes that resulted in 17 articles with 114 formulations. Eleven properties (input parameters) related to drugs and niosomes affecting particle size and drug entrapment (EE%) (output variables) were precisely identified and used for the network training. The network architecture consists of 5 fully connected hidden layers with eleven hidden nodes in each layer, input layer has eleven nodes where each of these nodes maintains omnidirectional injective relation with the corresponding hidden node. The hyperbolic tangent sigmoid transfer function (HTSTF) with Levenberg-Marquardt backpropagation (LMB) was used to train the model. The network showed the highest prediction accuracy of 93.76% and 91.79% for EE% and particle size prediction. Sensitivity analysis identified drug/lipid ratio and Cholesterol/Surfactant ratio as the most significant factors affecting EE% and particle size of niosomes. Accordingly, nine Donepezil hydrochloride (DNPZ) noisome batches were prepared using a 3x3 factorial design with drug/lipid ratio and cholesterol/surfactant ratio as factors to validate the developed model for a new drug. The model was able to reach a prediction accuracy of more than 97% for experimental batches. Finally, the superiority of global ANN was demonstrated compared to the local RSM model for DNPZ niosome formulations. In conclusion, this study has demonstrated the use of the data mining approach to group published scattered information and formulate a data set to train the ANN that can help to design an optimum formulation and processing parameters for a clinically viable medicine with minimum time, cost, and efforts. However, the key challenge in forming an efficient ANN lies in having a suitable large data set and using the correct algorithm for the network.
- Published
- 2021
- Full Text
- View/download PDF