PeakBot

PeakBot performs chromatographic peak picking in LC-HRMS profile-mode datasets to detect and classify chromatographic peaks for accurate feature detection and quantification in untargeted metabolomics.

Key Features:

• Local-Maxima Detection: Identifies local signal maxima within chromatograms as candidate peaks.
• Standardized Area Extraction: Converts each detected local maximum into a super-sampled standardized area defined in retention time (rt) versus mass-to-charge ratio (m/z).
• Machine Learning Integration: Applies a custom-trained convolutional neural network (CNN) implemented with TensorFlow 2.5.0 to classify standardized areas and provide peak center and bounding box.
• High Performance: Reports an accuracy of 0.99 on training and independent validation datasets for discriminating true peaks from background signals.
• Training Requirements: Requires a minimum of 100 reference features for model training.

Scientific Applications:

• Metabolic Feature Detection: Enables accurate detection and quantification of chromatographic peaks in LC-HRMS datasets.
• Untargeted Metabolomics Profiling: Facilitates comprehensive metabolite profiling and identification in untargeted metabolomics studies.

Methodology:

Identifies local maxima, transforms each into a super-sampled standardized rt × m/z area, and classifies these areas with a CNN implemented in TensorFlow 2.5.0 that outputs peak center and bounding box; model training uses at least 100 reference features and was evaluated on training and independent validation datasets achieving 0.99 accuracy.