Machine learning algorithms for prediction of entrapment efficiency in nanomaterials

Area of Science:

• Nanotechnology and Materials Science
• Computational Chemistry and Cheminformatics
• Pharmaceutical Sciences

Background:

• Drug entrapment efficiency in nanocarriers is crucial for therapeutic efficacy.
• Predicting this efficiency is complex due to multiple influencing factors, necessitating extensive experimental validation.
• Current comparative studies on predictive models for nanomaterial entrapment efficiency are limited.

Purpose of the Study:

• To develop and compare machine learning regression algorithms for predicting drug entrapment efficiency in nanomaterials.
• To identify the most effective algorithm for accurate and efficient prediction of entrapment efficiency.
• To determine the key formulation parameters influencing drug entrapment in niosomes.

Main Methods:

• Utilized supervised machine learning, specifically CatBoost, linear regression, support vector regression, and artificial neural networks.
• Trained and evaluated models using existing data to predict entrapment efficiency in niosomes.
• Performed comparative analysis of algorithm performance based on R² score and mean square error.

Main Results:

• The CatBoost algorithm achieved the highest performance, with an R² score of 0.98 and minimal mean square error (< 10⁻⁴).
• The drug:lipid ratio was identified as the most significant factor influencing entrapment efficiency.
• The lipid:surfactant molar ratio was found to be the second most important parameter.

Conclusions:

• Supervised machine learning, particularly CatBoost, offers a powerful tool for predicting drug entrapment efficiency in niosomes.
• This approach can significantly reduce experimental time and costs associated with nanosystem development.
• Optimizing drug:lipid and lipid:surfactant ratios using ML predictions can lead to enhanced nanosystem performance.