In this paper, we introduce an approach for analyzing complex biological data obtained from metabolomic analytical platforms. Such platforms generate massive and complex data that need appropriate methods for discovering meaningful biological information. The datasets to analyze consist in a limited set of individuals and a large set of attributes (variables). In this study, we are interested in mining metabolomic data to identify predictive biomarkers of metabolic diseases, such as type 2 diabetes. Our experiments show that a combination of numerical methods, e.g. SVM, Random Forests (RF), and ANOVA, with a symbolic method such as FCA, can be successfully used for discovering the best combination of predictive features. Our results show that RF and ANOVA seem to be the best suited methods for feature selection and discovery. We then use FCA for visualizing the markers in a suggestive and interpretable concept lattice. The outputs of our experiments consist in a short list of the 10 best potential predictive biomarkers.