peakonly

peakonly applies convolutional neural networks to detect and integrate chromatographic peaks in raw high-resolution LC-MS data for metabolomics and related GC- or LC-MS analyses.

Key Features:

• Convolutional Neural Networks: Uses CNNs to analyze raw LC-MS signal patterns for improved peak detection and integration.
• Dual Neural Network System: Employs a two-stage neural network approach with separate models for classification and boundary refinement.
• ROI Classification: The first network classifies regions of interest (ROIs) into noise, peaks, or uncertain peaks.
• Boundary Refinement and Area Integration: The second network determines precise peak boundaries to enable accurate peak area integration.
• High Precision and Noise Handling: Designed to maximize true positive peak identification while allowing detection or exclusion of low-intensity noisy peaks to reduce false positives.
• Adaptability Across Techniques: Applicable to high-resolution LC-MS workflows and adaptable to high-resolution GC- or LC-MS analyses.
• Implementation: Implemented in Python (version 3.5) and operates on raw LC-MS data.

Scientific Applications:

• Metabolomics: Improves peak detection and integration to support reliable identification and quantification of metabolites.
• Biomarker Discovery: Enhances detection sensitivity and precision to aid discovery of metabolite biomarkers.
• Metabolic Pathway Analysis: Provides accurate peak areas for downstream quantitative and pathway elucidation studies.
• High-Resolution GC-/LC-MS Studies: Can be adapted for peak detection and integration in other high-resolution GC- or LC-MS applications beyond metabolomics.

Methodology:

Train and apply two convolutional neural networks on raw LC-MS data: the first classifies ROIs as noise, peaks, or uncertain, and the second refines peak boundaries for accurate area integration.