PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Qualitative Evaluation of Chromatographic Data from Quality Control Schemes using a SVM
Montserrat Ventura, Antoni Sanchez-Niubo, Francisco Javier Ruiz, Núria Agell, Rosa Ventura, Cecilio Angulo, Antonia Domingo-Salvany, Jordi Segura and Rafael de la Torre
The Analyst Volume 132, Number 1, pp. 105-111, 2008. ISSN 0003-2654

Abstract

The qualitative evaluation of chromatographic data in the framework of external quality assurance schemes is considered in this paper. The homogeneity in the evaluation of chromatographic data among human experts in samples with analytes close to the limit of detection of analytical methods was examined and also a Support Vector Machine (SVM) was developed as an alternative to experts for a more homogeneous and automatic evaluation. A set of 105 ion chromatograms obtained by anti-doping control laboratories was used in this study. The quality of the ion chromatograms was evaluated qualitatively by nine independent experts (associating a score from 0 to 4) and also more objectively taking into account chromatographic parameters (peak width, asymmetry, resolution and S/N ratio). Results obtained showed a high degree of variability among experts when judging ion chromatograms. Experts applying extremely outlying evaluation criteria were identified and excluded from the data used to develop the SVM. This machine was built providing the system with qualitative information (scores assigned by experts) and with objective data (parameters) of the ion chromatograms. A seven-fold cross-validation approach was used to train and to evaluate the predictive ability of the machine. According to the results obtained, the SVM developed was found to be close to the reasoning process followed by the homogeneous human expert group. This machine also could provide a scoring system to sort laboratories according to the quality of their results. The qualitative evaluation of analytical records using a scoring system allowed the identification of the main factors affecting the quality of chromatographic analytical data, such as the specific analytical technique applied and the adherence to guidelines for reporting positive results.

EPrint Type:Article
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Learning/Statistics & Optimisation
ID Code:4532
Deposited By:Cecilio Angulo
Deposited On:13 March 2009