بيتا ياراحمدي

Molecular imprinted polymers are cross-linked polymers that show a selectivity pattern with respect to the shape, dimensions, type, and number of functional groups of the molecule. The practical synthesis and optimization needed to prepare these polymers has many disadvantages, including the need to spend time, spend money, the need to consume chemicals that are harmful to human health and the environment, the need for an operator to perform analyses, the need for equipment and laboratory equipment, etc. Imprinting factor is a quality eva‎luation measure of the imprinting. The higher the value of the immprinting factor, it means that the imprinting has been done with a higher and better quality. Machine learning is used to predict and categorize different variables. In this paper, using machine learning, we succeeded in creating a model with high accuracy to predict the imprinting factor of different polymers. For modeling, various feature selection methods including mutual information method, statistical correlation, dimensionality reduction, recurrent feature removal, forward selection, validation methods such as test-training and K-Fold validation method, as well as regression algorithms including seven algorithms Linear regression, congruence, elastic net, decision trees, Lasso, k nearest neighbor and four general algorithms like ADA, gradient boosting, random forest, and extra trees were used. As a result of using data cleaning and using RFE feature selection and gradient ensemble algorithm, we creat a model with an accuracy of %0.087. The use of machine learning increases speed and accuracy, reduces time, significantly reduces costs, does not require the use of chemical compounds, and does not require equipment. Finally, to check the efficiency of the prepared polymers, the figures of merit including accuracy, sensitivity, linear range, repeatability, selectivity and calibration curve for two different molecular template polymers (riboflavin and cephalexin-cadmium complex) were investigated. According to the results, polymers showed high accuracy and reproducibility for extracting target molecules.

يادگيري ماشين , الگوريتم‌هاي رگرسيون , پليمر قالب مولكولي , الگوريتم‌هاي تقويتي

machine learning , regression algorithms , molecularly imprinted polymer , ensemble algorithms.

bita yarahmadi

Dr. Seyed Mohammad Reza Milani Hosseini Dr. Seyed Majid Hashemianzadeh