VOCCluster

VOCCluster performs unsupervised clustering of volatile organic compound (VOC) features extracted from deconvolved Gas Chromatography Mass Spectrometry (GC/MS) breath data to group features by similar mass spectra and retention index for clinical breath metabolomics analysis.

Key Features:

• Unsupervised clustering: Implements an unsupervised clustering technique developed in Python for VOC feature grouping.
• Heuristic ontology: Employs a heuristic ontology derived from expert observations during data processing with various software packages.
• Input data: Operates on mass spectra and retention index profiles extracted from deconvolved GC/MS breath data.
• Preprocessing and scaling: Addresses processing, aligning, scaling, and clustering of thousands of features extracted from GC/MS data.
• Benchmarking: Evaluated against known ground truth compounds and compared to DBSCAN and OPTICS clustering methods.
• Performance: Achieved 96% accuracy (±0.04 at a 95% confidence interval) in clustering features into 1081 distinct groups.
• Group composition: Clusters encompass endogenous and exogenous compounds as well as instrumental artifacts.
• Clinical dataset: Applied to breath samples from 74 participants, clustering over 15,000 features collected before and after radiation therapy.

Scientific Applications:

• Clinical breath GC/MS analysis: Groups VOC features to support breath-based clinical metabolomics studies.
• Metabolic profiling and diagnostics: Accelerates metabolic profiling and advances discovery platforms for next-generation medical diagnostic applications.
• Method benchmarking in metabolomics: Provides comparative evaluation of clustering approaches using ground truth and comparisons to DBSCAN and OPTICS.
• Compound and artifact discrimination: Enables separation of endogenous and exogenous compounds from instrumental artifacts in breath data.

Methodology:

Implements an unsupervised clustering algorithm developed in Python using a heuristic ontology derived from expert observations, operating on mass spectra and retention index profiles from deconvolved GC/MS breath data and including processing, alignment, scaling and clustering of thousands of features with evaluation against known ground truth and comparison to DBSCAN and OPTICS.