PHYSICAL PROPERTY PREDICTION MODEL FOR PREDICTING MELTING POINT BASED ON QUANTITATIVE STRUCTURE-ACTIVITY RELATIONSHIP USING LINEAR AND NONLINEAR MACHINE LEARNING METHODS

摘要

The present invention relates to a physical property prediction model for predicting a melting point based on a quantitative structure-activity relationship using linear and nonlinear machine learning methods. According to the present invention, the physical property prediction model for predicting a melting point based on a quantitative structure-activity relationship using linear and nonlinear machine learning methods comprises the steps of: (1) collecting and refining compound data having a melting point value; (2) collecting and calculating a descriptor that can be calculated in a molecular structure of the compound data; (3) dividing the collected and refined compound data into a training set and an external validation set; (4) performing pre-filtering of the descriptor; (5) selecting a descriptor suitable for prediction of a melting point; (6) increasing a value of the descriptor, accumulating respective optimized machine learning modules and results, generating combined machine learning models reducing an error generated in the respective optimized machine learning modules through machine learning models generated by combining two or more among calculated machine learning modules, comparing the combined machine learning models with one another, and determining a final combined machine learning model; (7) setting and applying an applicable range for evaluation of reliability in the final combined machine learning model; and (8) determining the reliability and suitability of the final combined machine learning model.;COPYRIGHT KIPO 2016