Search academic texts

Information × Registration Number 0222U005458, 0121U100646 , R & D reports Title QSPR models for predicting the main criteria of similarity of real gases and recovery of elasticity curves of organic compounds which have not been studied experimentally popup.stage_title Head Kuz’min Victor Ye., Доктор хімічних наук Registration Date 30-12-2022 Organization Physico-Chemical Institute O. Bogatsky National Academy of Sciences of Ukraine popup.description2 The object of research is a system of structural description of organic compounds for prediction of the main criteria of similarity of real gases (critical compressibility factor (Zc), Pitzer's acentricity factor (ω) and Riedel's criterion (αc)), structural and physicochemical factors influencing on the studied properties. The aim of the work is to develop QSPR models for prediction of the main criteria of similarity of real gases and recovery of elasticity curves of organic compounds of different structure for the development of chemical-technological processes. Research methods - QSPR modeling (Quantitative Structure-Property Relationships - quantitative relations "structure-property") based on the method of simplex representation of molecular structure; statistical methods of linear regression analysis, partial least squares method (PLS), trend vector method, random forest method, support vector machine. Object of development - QSPR models for predicting the main criteria of similarity of real gases. At the second stage of the work adequate 1D-2D QSPR-models has been developed on the base of training samples formed at the previous stage for the main criteria of similarity to real gases. Simplex and Dragon descriptors (topological indices, fragmentary descriptors, molecular structure characteristics) in combination with the random forest statistical method were used to build statistical models. This method actually represents an ensemble of decision trees and allows building fairly high-quality consensus models. At the modeling stage various types of descriptors were varied those that were mutually correlated were screened out, and the most important descriptors were selected using the trend-vector procedure. Models with high quality of approximation and high quality of prediction within the out-of-back procedure were selected. Domain applicability were determined for all models; the error of prediction is comparable to the experimental one. Product Description popup.authors Illyushko Natalia О. Artemenko Anatoliy G. Kuz’min Victor Ye. Ognichenko Liudmila M. Khromov Olexander I. popup.nrat_date 2022-12-30 Close