These tools will no longer be maintained as of December 31, 2024. Archived website can be found here. PubMed4Hh GitHub repository can be found here. Contact NLM Customer Service if you have questions.


Pubmed for Handhelds

PUBMED FOR HANDHELDS

Search MEDLINE/PubMed

  • Title: Cross-column prediction of gas-chromatographic retention of polychlorinated biphenyls by artificial neural networks.
    Author: D'Archivio AA, Incani A, Ruggieri F.
    Journal: J Chromatogr A; 2011 Dec 02; 1218(48):8679-90. PubMed ID: 22000780.
    Abstract:
    In this paper, we build a multiple-column retention model able to predict the behaviour of polychlorinated biphenyls (PCBs) in capillary gas-chromatography (GC) within a wide range of separation conditions. To this end, GC retention is related to both chemical structure of PCBs, encoded by selected theoretical molecular descriptors, and the kind of stationary phase, represented by the relative retention time (RRT) of a suitable small number of analytes. The model was generated using the retention data of 70 PCBs extracted from the pool of the 209 possible congeners collected on 17 different capillary columns featured by non-polar or moderately polar stationary phases, reported in the literature. Multilinear regression combined with genetic algorithm variable selection was preliminarily applied to generate a four-dimensional quantitative structure-retention relationship (QSRR) for each of the 17 columns, based on theoretical molecular descriptors extracted from the large set provided by the software Dragon. 33 molecular descriptors obtained by merging the non-common descriptors of various single-column QSRRs, combined with RRTs values of the less and the most retained PCB, were considered as the starting independent variables of the multiple-column retention model. A multi-layer artificial neural network (ANN), optimised on a validation set extracted from the calibration data, was applied to generate the multi-column retention model. The influence of starting inputs on the network output was evaluated by a sensitivity analysis and model complexity was reduced through a step-wise elimination of redundant molecular descriptors, while RRTs of further PCBs were included to improve description of the stationary phase. Nine molecular descriptors and RRTs of eight selected PCBs are considered as the independent variables of the final ANN-based model, whose predictive performance was tested on the 139 PCBs excluded from calibration and on six external columns and/or temperature programs.
    [Abstract] [Full Text] [Related] [New Search]