Abstract
Basic chemometric methods for making empirical regression models for QSPR/QSAR are briefly described from a user's point of view. Emphasis is given to PLS regression, simple variable selection and a careful and cautious evaluation of the performance of PLS models by repeated double cross validation (rdCV). A demonstration example is worked out for QSPR models that predict gas chromatographic retention indices (values between 197 and 504 units) of 209 polycyclic aromatic compounds (PAC) from molecular descriptors generated by Dragon software. Most favorable models were obtained from data sets containing also descriptors from 3D structures with all H-atoms (computed by Corina software), using stepwise variable selection (reducing 2688 descriptors to a subset of 22). The final QSPR model has typical prediction errors for the retention index of +12 units (95% tolerance interval, for test set objects). Programs and data are provided as supplementary material for the open source R software environment.
Highlights
Properties or activities of chemical compounds can be estimated The principal aims of model building are small prediction errors |y - by regression models that relate a set of molecular descriptors with the ŷ| for data from substances in a test set that have not been used property or activity
Emphasis is given to PLS regression, simple variable selection and a careful and cautious evaluation of the performance of PLS models by repeated double cross validation
A QSPR demonstration example has been worked out using a set of partially new R functions for molecular descriptor data as generated by Dragon software
Summary
Properties or activities of chemical compounds can be estimated The principal aims of model building are small prediction errors |y - by regression models that relate a set of molecular descriptors with the ŷ| for data from substances in a test set that have not been used property or activity. In the case of calibration models (modeling/predicting a continuous property y) the residuals (prediction errors), ei approaches have been published and implemented into the software, so it is less strictly defined than the traditional method OLS (ordinary ei = yi - ŷi least-squares regression, mostly not directly applicable to chemistryrelated data because of correlating variables and a larger number of variables than number of objects), and PCR (principal component regression, similar to PLS).
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callMore From: Computational and Structural Biotechnology Journal
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
