Outlier feature classification

outlier-feature-classification

License: restricted — no clear open-source license detected for the underlying tool; verify licensing before commercial use or redistribution. <!-- asb-license-banner -->

Summary

Identify and classify molecular features with anomalous mass-to-charge (m/z) decimal values that fall outside the expected instrumental range, removing them from downstream analysis in untargeted LC-MS/MS metabolomics. This filtering step ensures that only features with physically plausible m/z values are retained for metabolite annotation and statistical testing.

When to use

After peak picking by MS-DIAL and import into R, when the feature table contains m/z values with decimal components that fall within the [4, 8] interval (indicating instrumental artifacts or calibration errors). Apply this filter during the initial quality-control and filtering phase, before metabolite annotation or statistical analysis, to prevent false-positive identifications and spurious associations.

When NOT to use

• When the input data has already been filtered for m/z quality by the peak-picking software or a prior preprocessing step.
• When analyzing targeted metabolomics data with a predefined list of m/z values, where all features are known to be biologically relevant.
• When the m/z decimal range [4, 8] is not representative of instrumental artifacts in your specific MS-DIAL configuration or ionization mode.

Inputs

• MS-DIAL feature table with m/z values (numeric column)
• feature abundance matrix (features × samples)

Outputs

• filtered feature table (appropriate m/z values only)
• outlier feature table (inappropriate m/z values)
• filtering summary report (count of retained vs. removed features)

How to apply

Extract the decimal part of each m/z value by computing m/z modulo 1 for all features in the MS-DIAL feature table. Classify features as inappropriate if their decimal part lies within the interval [4, 8]; otherwise mark as appropriate. Separate the feature set into two subsets: one containing features with appropriate m/z values (suitable for downstream analysis) and one containing features with inappropriate m/z values (to be excluded). Return both subsets and a summary report documenting the count and proportion of filtered features. The rationale is that features with m/z decimal values in this range typically reflect instrumental noise or calibration drift rather than true metabolite signals.

Related tools

• margheRita (Implements m/z filtering via the m_z_filtering() function within the complete LC-MS/MS preprocessing workflow) — https://github.com/emosca-cnr/margheRita
• MS-DIAL (Generates the initial feature table with m/z values that serves as input to outlier classification) — http://prime.psc.riken.jp/Metabolomics_Software/MS-DIAL/
• R (Programming environment for implementing decimal extraction and feature classification logic)
• notame (Alternative R package for LC-MS metabolomics preprocessing that includes filtering by wrong m/z values) — https://github.com/hanhineva-lab/notame

Evaluation signals

• Verify that the sum of retained and removed feature counts equals the total input feature count.
• Confirm that all features in the 'appropriate' subset have m/z decimal values outside [4, 8] (i.e., decimal part < 4 OR decimal part > 8).
• Confirm that all features in the 'inappropriate' subset have m/z decimal values within [4, 8] (i.e., 4 ≤ decimal part ≤ 8).
• Check that the filtered feature table (appropriate subset) yields improved cosine similarity matches or higher confidence metabolite identifications in downstream annotation steps compared to the unfiltered table.
• Ensure the feature abundance matrix dimensions are consistent after subsetting (number of columns unchanged; number of rows equals count of retained features).

Limitations

• The m/z decimal interval [4, 8] is a default threshold specific to margheRita and MS-DIAL configuration; it may need adjustment for different mass spectrometers, ionization modes (RP-C18, HILIC, RP-C8, pZIC-HILIC), or calibration protocols.
• This filter assumes that features with m/z decimal values in [4, 8] are uniformly artifacts; true signals falling in this range (though rare) will be incorrectly removed.
• The method does not account for m/z shifts caused by adduct formation, isotopologues, or in-source fragmentation; these must be handled by separate preprocessing steps.
• Filtering is applied before consideration of other quality metrics (e.g., coefficient of variation, missing value patterns), so the final dataset may still contain low-quality features after this step.

Evidence

• [other] The m/z filtering function removes features whose m/z decimal values fall within the interval [4, 8] by default. When applied to the dataset, this filter excluded 56 features with inappropriate m/z values while retaining 548 features with appropriate m/z values.: "The m/z filtering function removes features whose m/z decimal values fall within the interval [4, 8] by default. When applied to the dataset, this filter excluded 56 features with inappropriate m/z"
• [other] Extract the decimal part of each m/z value by computing m/z modulo 1. Classify features as inappropriate if the decimal part lies within [4,8]; otherwise mark as appropriate.: "Extract the decimal part of each m/z value by computing m/z modulo 1. Classify features as inappropriate if the decimal part lies within [4,8]; otherwise mark as appropriate."
• [readme] The package provides a series of pre-processing functions (quality control, filtering and normalization) with a particular focus on methods specifically recommended for metabolomic profiles, such as filtering by mass defects, filtering by coefficient of variation (samples vs QCs) and probabilistic quotient normalization: "a series of pre-processing functions (quality control, filtering and normalization) with a particular focus on methods specifically recommended for metabolomic profiles, such as filtering by mass"
• [intro] runs filters to exclude features/sample with many missing values, features with wrong m/z values: "runs filters to exclude features/sample with many missing values, features with wrong m/z values"