biosigner

biosigner identifies minimal subsets of discriminant molecular features from high-throughput omics datasets such as transcriptomics and metabolomics to support biomarker discovery and molecular signature definition for diagnostic and clinical validation.

Key Features:

• Feature Selection Methodology: Employs resampling, ranking of variable importance, significance assessment via permutation of feature values in test subsets, and half-interval search to select the smallest subset of features that significantly contribute to model performance.
• Integration with Binary Classifiers: Operates with Partial Least Squares-Discriminant Analysis (PLS-DA), Random Forest, and Support Vector Machines (SVM) as the underlying binary classifiers.
• Performance and Efficiency: Applied to biological and clinical datasets containing up to 7000 features, it generates complementary signatures that often yield higher prediction accuracies than initial full models and completes selection in minutes.
• Comparison with Alternative Approaches: Produces smaller and more stable molecular signatures compared with alternative feature-selection methods.
• Clinical Application Example: Distinguished type 1 versus type 2 diabetic patients in metabolomic data, identifying discriminating features including a fragment of taurochenodeoxycholic bile acid.

Scientific Applications:

• Biomarker discovery in omics: Facilitates identification of robust molecular signatures from transcriptomics and metabolomics datasets for downstream validation.
• Diagnostic signature development and validation: Supports transition from untargeted biomarker discovery to targeted validation phases aimed at developing clinically relevant diagnostic tests.

Methodology:

Resampling, ranking of variable importance, permutation-based significance assessment on test subsets, half-interval search for feature subset selection, and integration with PLS-DA, Random Forest, and SVM classifiers.