A new approach to prediction riboflavin absorbance using imprinted polymer and ensemble machine learning algorithms

Area of Science:

• Analytical Chemistry
• Polymer Science
• Computational Chemistry

Background:

• Molecularly imprinted polymers (MIPs) are effective for quantifying riboflavin (vitamin B2) via UV/Vis spectrophotometry.
• Optimizing MIP synthesis conditions is resource-intensive, requiring time, money, chemicals, and laboratory equipment.
• Machine learning (ML) offers a novel approach to predict optimal synthesis parameters and maximize riboflavin absorption.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting riboflavin absorbance.
• To identify key factors influencing MIP performance for riboflavin detection.
• To demonstrate the efficiency of ML-driven optimization over traditional laboratory methods.

Main Methods:

• Synthesized MIPs for selective riboflavin extraction and quantified absorbance using UV/Vis spectrophotometry.
• Investigated the impact of six factors: template molar ratio, monomer molar ratio, cross-linker molar ratio, loading time, stirring rate, and pH.
• Employed ensemble ML algorithms (Gradient Boosting, Extra Trees, Random Forest, AdaBoost) and mutual information for feature selection.

Main Results:

• Mutual information analysis identified template, monomer, and cross-linker molar ratios as highly influential factors.
• Gradient Boosting (GB) and AdaBoost algorithms outperformed Extra Trees and Random Forest.
• The optimized GB model (n-estimator=300) achieved high accuracy, with R²=0.966, MAE=-0.0037, and MSE=-0.000078.

Conclusions:

• Machine learning models can accurately predict riboflavin absorbance, theoretically determining optimal synthesis conditions.
• The proposed ML approach significantly reduces the need for extensive laboratory work, saving time, cost, and materials.
• This study highlights the potential of computational methods to streamline the development and optimization of analytical tools like MIPs.