Molecular Descriptor Subset Selection in Theoretical Peptide Quantitative Structure鈥揜etention Relationship Model Development Using Nature-Inspired Optimization Algorithms

详细信息    查看全文

• 作者：Petar 沤uvela ; J. Jay Liu ; Katarzyna Macur ; Tomasz B膮czek
• 刊名：Analytical Chemistry
• 出版年：2015
• 出版时间：October 6, 2015
• 年：2015
• 卷：87
• 期：19
• 页码：9876-9883
• 全文大小：407K
• ISSN：1520-6882

文摘

In this work, performance of five nature-inspired optimization algorithms, genetic algorithm (GA), particle swarm optimization (PSO), artificial bee colony (ABC), firefly algorithm (FA), and flower pollination algorithm (FPA), was compared in molecular descriptor selection for development of quantitative structure鈥搑etention relationship (QSRR) models for 83 peptides that originate from eight model proteins. The matrix with 423 descriptors was used as input, and QSRR models based on selected descriptors were built using partial least squares (PLS), whereas root mean square error of prediction (RMSEP) was used as a fitness function for their selection. Three performance criteria, prediction accuracy, computational cost, and the number of selected descriptors, were used to evaluate the developed QSRR models. The results show that all five variable selection methods outperform interval PLS (iPLS), sparse PLS (sPLS), and the full PLS model, whereas GA is superior because of its lowest computational cost and higher accuracy (RMSEP of 5.534%) with a smaller number of variables (nine descriptors). The GA-QSRR model was validated initially through Y-randomization. In addition, it was successfully validated with an external testing set out of 102 peptides originating from Bacillus subtilis proteomes (RMSEP of 22.030%). Its applicability domain was defined, from which it was evident that the developed GA-QSRR exhibited strong robustness. All the sources of the model鈥檚 error were identified, thus allowing for further application of the developed methodology in proteomics.