Deep Learning-Assisted Peak Curation for Large-Scale LC-MS Metabolomics.

Area of Science:

• Computational Biology
• Analytical Chemistry
• Biotechnology

Background:

• Untargeted metabolomics is crucial for biological discovery.
• Existing automated peak detection methods in LC-MS often lack precision, leading to false positives.
• Accurate peak identification is essential for robust data analysis.

Purpose of the Study:

• To introduce NeatMS, a novel machine learning tool for precise peak detection in untargeted metabolomics.
• To reduce the number and fraction of false peaks in LC-MS data.
• To enhance the scalability and robustness of metabolomic data analysis.

Main Methods:

• Development of NeatMS utilizing a convolutional neural network (CNN) for peak detection.
• Implementation of a pre-trained model incorporating expert knowledge for signal-to-noise differentiation.
• Facilitation of model training and transfer learning for customized analysis.
• Integration into various LC-MS workflows and performance evaluation.

Main Results:

• NeatMS significantly reduces false positive peaks compared to existing methods.
• The tool demonstrates high precision in differentiating true chemical signals from noise.
• Successful integration and validation across different LC-MS analysis pipelines.
• Performance comparison highlights NeatMS's superiority over other approaches.

Conclusions:

• NeatMS offers a robust and scalable solution for accurate peak detection in untargeted metabolomics.
• The machine learning-based approach enhances data quality and reliability for large-scale studies.
• Open-source availability and accessible packages (PyPi, Bioconda) promote widespread adoption.