H Blasco1, J Błaszczyński2, J C Billaut3, L Nadal-Desbarats4, P F Pradat5, D Devos6, C Moreau6, C R Andres7, P Emond4, P Corcia8, R Słowiński9. 1. Inserm U930, Tours, France; Université François-Rabelais, Tours, France; Laboratoire de Biochimie et Biologie Moléculaire, CHRU de Tours, Tours, France. Electronic address: helene.blasco@univ-tours.fr. 2. Institute of Computing Science, Poznań University of Technology, 60-965 Poznań, Poland. 3. Université François-Rabelais de Tours, CNRS, LI EA 6300, OC ERL CNRS 6305, Tours, France. 4. Inserm U930, Tours, France; Université François-Rabelais, Tours, France; PPF, Université François-Rabelais, Tours, France. 5. Fédération des Maladies du Système Nerveux, Centre Référent Maladie Rare SLA, Hôpital de la Pitié-Salpétrière, Paris, France. 6. Service de Neurologie, CHRU de Lille, Lille, France. 7. Inserm U930, Tours, France; Université François-Rabelais, Tours, France; Laboratoire de Biochimie et Biologie Moléculaire, CHRU de Tours, Tours, France. 8. Inserm U930, Tours, France; Université François-Rabelais, Tours, France; Centre SLA, Service de Neurologie, CHRU Bretonneau, Tours, France. 9. Institute of Computing Science, Poznań University of Technology, 60-965 Poznań, Poland; Systems Research Institute, Polish Academy of Sciences, 01-447 Warsaw, Poland.
Abstract
BACKGROUND: Metabolomics is an emerging field that includes ascertaining a metabolic profile from a combination of small molecules, and which has health applications. Metabolomic methods are currently applied to discover diagnostic biomarkers and to identify pathophysiological pathways involved in pathology. However, metabolomic data are complex and are usually analyzed by statistical methods. Although the methods have been widely described, most have not been either standardized or validated. Data analysis is the foundation of a robust methodology, so new mathematical methods need to be developed to assess and complement current methods. We therefore applied, for the first time, the dominance-based rough set approach (DRSA) to metabolomics data; we also assessed the complementarity of this method with standard statistical methods. Some attributes were transformed in a way allowing us to discover global and local monotonic relationships between condition and decision attributes. We used previously published metabolomics data (18 variables) for amyotrophic lateral sclerosis (ALS) and non-ALS patients. RESULTS: Principal Component Analysis (PCA) and Orthogonal Partial Least Square-Discriminant Analysis (OPLS-DA) allowed satisfactory discrimination (72.7%) between ALS and non-ALS patients. Some discriminant metabolites were identified: acetate, acetone, pyruvate and glutamine. The concentrations of acetate and pyruvate were also identified by univariate analysis as significantly different between ALS and non-ALS patients. DRSA correctly classified 68.7% of the cases and established rules involving some of the metabolites highlighted by OPLS-DA (acetate and acetone). Some rules identified potential biomarkers not revealed by OPLS-DA (beta-hydroxybutyrate). We also found a large number of common discriminating metabolites after Bayesian confirmation measures, particularly acetate, pyruvate, acetone and ascorbate, consistent with the pathophysiological pathways involved in ALS. CONCLUSION: DRSA provides a complementary method for improving the predictive performance of the multivariate data analysis usually used in metabolomics. This method could help in the identification of metabolites involved in disease pathogenesis. Interestingly, these different strategies mostly identified the same metabolites as being discriminant. The selection of strong decision rules with high value of Bayesian confirmation provides useful information about relevant condition-decision relationships not otherwise revealed in metabolomics data.
BACKGROUND: Metabolomics is an emerging field that includes ascertaining a metabolic profile from a combination of small molecules, and which has health applications. Metabolomic methods are currently applied to discover diagnostic biomarkers and to identify pathophysiological pathways involved in pathology. However, metabolomic data are complex and are usually analyzed by statistical methods. Although the methods have been widely described, most have not been either standardized or validated. Data analysis is the foundation of a robust methodology, so new mathematical methods need to be developed to assess and complement current methods. We therefore applied, for the first time, the dominance-based rough set approach (DRSA) to metabolomics data; we also assessed the complementarity of this method with standard statistical methods. Some attributes were transformed in a way allowing us to discover global and local monotonic relationships between condition and decision attributes. We used previously published metabolomics data (18 variables) for amyotrophic lateral sclerosis (ALS) and non-ALS patients. RESULTS: Principal Component Analysis (PCA) and Orthogonal Partial Least Square-Discriminant Analysis (OPLS-DA) allowed satisfactory discrimination (72.7%) between ALS and non-ALS patients. Some discriminant metabolites were identified: acetate, acetone, pyruvate and glutamine. The concentrations of acetate and pyruvate were also identified by univariate analysis as significantly different between ALS and non-ALS patients. DRSA correctly classified 68.7% of the cases and established rules involving some of the metabolites highlighted by OPLS-DA (acetate and acetone). Some rules identified potential biomarkers not revealed by OPLS-DA (beta-hydroxybutyrate). We also found a large number of common discriminating metabolites after Bayesian confirmation measures, particularly acetate, pyruvate, acetone and ascorbate, consistent with the pathophysiological pathways involved in ALS. CONCLUSION: DRSA provides a complementary method for improving the predictive performance of the multivariate data analysis usually used in metabolomics. This method could help in the identification of metabolites involved in disease pathogenesis. Interestingly, these different strategies mostly identified the same metabolites as being discriminant. The selection of strong decision rules with high value of Bayesian confirmation provides useful information about relevant condition-decision relationships not otherwise revealed in metabolomics data.
Authors: Jonathan James Grandy; Bora Onat; Verena Tunnicliffe; David A Butterfield; Janusz Pawliszyn Journal: Sci Rep Date: 2020-01-28 Impact factor: 4.379
Authors: Alessandra Tata; Ivana Pallante; Andrea Massaro; Brunella Miano; Massimo Bottazzari; Paola Fiorini; Mauro Dal Prà; Laura Paganini; Annalisa Stefani; Jeroen De Buck; Roberto Piro; Nicola Pozzato Journal: Front Vet Sci Date: 2021-01-20
Authors: Vittoria Mallia; Steven Verhaegen; Bjarne Styrishave; Gunnar Sundstøl Eriksen; Malene Louise Johannsen; Erik Ropstad; Silvio Uhlig Journal: PLoS One Date: 2020-12-15 Impact factor: 3.240