wsPrediction

wsPrediction predicts chemical sub-groups within GC-MS mass spectrometry datasets by applying decision tree-based supervised machine learning to Golm Metabolome Database-derived structural features for identification of mass spectral tags (MSTs).

Key Features:

• GC-MS compatibility: Operates on gas chromatography–mass spectrometry (GC-MS) datasets and mass spectral tags (MSTs).
• Golm Metabolome Database integration: Leverages reference compound information from the Golm Metabolome Database (GMD) to derive structural features.
• Decision tree-based supervised learning: Uses decision tree (DT)-based supervised machine learning to classify chemical sub-groups.
• Structural feature extraction: Extracts structural features from GMD to subdivide the metabolite space and define prediction target classes.
• Mass spectral and RI features: Utilizes mass spectral features together with retention index (RI) information for substructure detection.
• Decision tree availability: Provides decision trees for inspection and batch processing through SOAP-based web services.
• Matching and structure search: Supports matching and structure search functionalities via SOAP-based web services.
• REST read-only access: Exposes database entities via a REST-compliant XML + HTTP interface for read-only access.
• Integration into spectral library: Integrates prediction decision trees into the GMD mass spectral library.
• MST identification without pure standards: Addresses identification and classification of MSTs in the absence of authenticated pure reference substances.

Scientific Applications:

• Large-scale biomarker screening: Enables large-scale screening for novel metabolic biomarkers in GC-MS datasets.
• MST classification: Facilitates classification and tentative identification of unidentified mass spectral tags (MSTs).
• Substructure detection: Detects chemical substructures within compounds using mass spectral and RI information.
• Library augmentation: Augments mass spectral libraries with decision tree–based substructure predictions.

Methodology:

Structural feature extraction from Golm Metabolome Database (GMD) reference compounds defines prediction target classes; supervised decision tree (DT)-based machine learning is applied using mass spectral features and retention index (RI) information; decision trees are made available for batch processing, matching and structure search via SOAP-based web services, with read-only access to database entities via a REST-compliant XML + HTTP interface, and integrated into the GMD mass spectral library.