Peak Detection In Lcms Data

peak-detection-in-lcms-data

Summary

Automated identification of chromatographic peaks across retention-time-corrected LC-MS samples to construct a feature table with m/z, retention time, and intensity values. This skill detects individual metabolic features for downstream annotation and statistical analysis in untargeted metabolomics workflows.

When to use

Apply this skill after retention-time correction has been completed on centroided LC-MS data (mzML or mzXML format) and you need to identify all detected peaks as a unified feature table across a cohort of samples. Requires at least one quality control (QC) file to have been designated during data import and alignment.

When NOT to use

• Input is already a feature table or integrated peak list — peak detection has already been performed.
• Raw data has not undergone retention-time correction or data alignment — peak detection requires aligned ROI matrices.
• No centroided LC-MS data available — peak detection expects centroid spectrum format, not profile mode.

Inputs

• Retention-time-corrected LC-MS raw data (mzML or mzXML format)
• Designation of at least one quality control (QC) file
• Optional: targeted extraction table (xlsx format) for targeted peak detection

Outputs

• Feature table with m/z, retention time, and intensity values
• Feature detection results in tabular format (CSV or mzTab)
• Chromatogram visualizations per detected feature
• Feature data export (xlsx or internal .pkd format for session persistence)

How to apply

Peak detection operates on aligned raw data within a Regions of Interest (ROI) matrix constructed from QC file signals. Set the m/z deviation tolerance (delta m/z, default 0.01) to control mass resolution; use minimum continuous non-zero points (default 3, increase for high-noise small-scale experiments) and maximum peak width (default 15 seconds) to filter noise and resolve coelution. Specify the minimum number of non-zero peaks across samples (default 80%) to avoid spurious features, and set intensity threshold (default 50) to exclude low-abundance noise. Select a reference QC file to represent feature apex retention time and m/z. Use the entropy coefficient (default 0.8, lower for large-scale studies) to control feature scoring stringency. Peak detection integrates centWave-based ROI detection with dynamic boundary identification to achieve low false positive and false negative rates simultaneously.

Related tools

• MetCohort (Integrated LC-MS data processing platform that executes peak detection using a robust centWave-based algorithm with ROI matrix construction and dynamic boundary identification for low false positive/negative rates) — https://github.com/JunYang2021/MetCohort

Evaluation signals

• Feature table contains no missing m/z, retention time, or intensity values for detected peaks across all samples.
• Feature intensity values fall within the specified minimum intensity threshold (default ≥50) and non-zero peaks meet the minimum sample requirement (default ≥80% of samples).
• Retention time and m/z of each feature match the apex position in the reference QC file used for feature representation.
• Chromatogram visualizations show detected peaks falling within the maximum peak width constraint (default ≤15 seconds) with shaded integration regions aligned to the specified ROI boundaries.
• Feature table can be exported and re-imported (.pkd format) without data loss, and export to xlsx format is valid and complete.

Limitations

• Peak detection quality depends critically on data alignment quality; poor retention-time alignment reduces feature identification and integration accuracy (confirmed by comparison of alignment deviation plots).
• For small-scale sample analyses with few QC files, increasing minimum continuous non-zero points is necessary to reduce noise-driven false features.
• Isomeric and coeluting compounds may have reduced detection accuracy in untargeted mode; targeted extraction table can improve specificity but with trade-off in integration precision.
• Abnormal files with noisy or significantly different chromatographic conditions may cause LOWESS fitting warnings and incorrect local matching; manual inspection of retention-time deviation plots is required for large-scale studies.
• Feature representation (retention time and m/z) is fixed to the selected reference QC file; changing the reference file changes feature representation and may complicate inter-table feature matching.

Evidence

• [other] Peak detection is performed as a sequential step after data correction, and the resulting feature detection outputs are visualized in the software and can be saved for subsequent analysis.: "Peak detection is performed as a sequential step after data correction, and the resulting feature detection outputs are visualized in the software and can be saved for subsequent analysis."
• [intro] With innovative and robust data correction and feature detection algorithm, MetCohort have a low false positive and false negative rate simultaneously. Feature table of high quality is generated: "With innovative and robust data correction and feature detection algorithm, MetCohort have a low false positive and false negative rate simultaneously. Feature table of high quality is generated"
• [readme] At least one file need to be specified as quality control (QC) file. Then data correction and peak detection should be performed in order.: "At least one file need to be specified as quality control (QC) file. Then data correction and peak detection should be performed in order."
• [readme] Delta m/z: Allowed m/z deviation in the process of ROI detection and ROI matrix construction. Default value is 0.01.: "Delta m/z: Allowed m/z deviation in the process of ROI detection and ROI matrix construction. Default value is 0.01."
• [readme] Minimum number of continuous non-zero points: Allowed minimum number of continuous non-zero points in the process of ROI detection with centWave algorithm. Default value is 3. In small-scale sample processing with a few QC files, the value should be increased to decrease noise.: "Minimum number of continuous non-zero points in the process of ROI detection with centWave algorithm. Default value is 3. In small-scale sample processing with a few QC files, the value should be"
• [readme] Maximum width of peak: Allowed maximum chromatographic width of detected features. Default value is 15 seconds.: "Maximum width of peak: Allowed maximum chromatographic width of detected features. Default value is 15 seconds."
• [readme] Minimum intensity of a feature: Allowed minimum intensity of a feature. If the maximum height of a feature in all the samples is lower than the value, the feature is filtered. Default value is 50.: "Minimum intensity of a feature: Allowed minimum intensity of a feature. If the maximum height of a feature in all the samples is lower than the value, the feature is filtered. Default value is 50."
• [readme] The retention time and m/z of one feature is represented by the retention time and m/z of the apex of the peak in the reference file.: "The retention time and m/z of one feature is represented by the retention time and m/z of the apex of the peak in the reference file."
• [readme] Entropy coefficient: A value ranging from 0 to 1 that controls the score threshold for feature determination in the ROI matrix. A larger entropy coefficient corresponds to a higher threshold. The default value is set to 0.8. For large-scale sample processing, users are advised to adjust this value lower.: "Entropy coefficient: A value ranging from 0 to 1 that controls the score threshold for feature determination in the ROI matrix. A larger entropy coefficient corresponds to a higher threshold. The"
• [readme] The quality of the alignment can significantly influence feature detection and integration. Therefore, it is necessary to manually inspect the alignment results during processing.: "The quality of the alignment can significantly influence feature detection and integration. Therefore, it is necessary to manually inspect the alignment results during processing."
• [readme] Good alignment can make the true features being more easily identified and integrated. A bad data alignment can reduce the feature numbers and negatively affect feature integration.: "Good alignment can make the true features being more easily identified and integrated. A bad data alignment can reduce the feature numbers and negatively affect feature integration."