Ms1 feature extraction

ms1-feature-extraction

Summary

Extract and integrate multiple MS1 chemical predictors (m/z, retention times, isotopic patterns, neutral loss patterns) from LC-HRMS features to compute unified proximity scores between detected features and suspect compounds. This enables rapid prioritization of compound candidates during suspect screening workflows.

When to use

When you have an LC-HRMS feature table (with m/z values, retention times, and isotopic signatures) and a suspect compound database (with reference m/z, expected retention time windows, isotope ratios, and neutral loss fragments), and you need to rapidly prioritize which features are most likely to correspond to which suspects before MS/MS annotation or manual review.

When NOT to use

• Input is already a manually validated or MS/MS-confirmed compound assignment — use this skill for pre-annotation only, not to second-guess confirmed identities.
• You lack reference isotopic or neutral loss patterns for your suspect compounds — the scoring will be incomplete and reliability will degrade.
• Your feature table does not include retention time data — RT proximity scoring will be unavailable, reducing discriminatory power.

Inputs

• LC-HRMS feature table (m/z values, retention times, isotopic pattern signatures)
• Suspect compound database (reference m/z, expected retention time windows, isotope ratios, neutral loss fragments)
• Mass error tolerance (ppm or absolute Da)
• Retention time deviation window (minutes)
• Component weights for ensemble scoring

Outputs

• Feature–suspect proximity scores (per feature–suspect pair)
• Ranked candidate compound list (sorted by proximity score)
• Filtered prioritized compound assignments (above score threshold)

How to apply

Load the feature table and suspect compound database into Scannotation. For each feature–suspect pair, compute four component scores: (1) m/z distance using absolute or relative mass error tolerance (typically ppm-based); (2) retention time proximity by comparing observed vs. predicted RT within an acceptable deviation window; (3) isotopic pattern match by comparing observed isotope abundance ratios to theoretical patterns for the suspect; (4) neutral loss pattern match by identifying common loss fragments (H₂O, CO₂, CH₄) in the observed feature fragmentation profile relative to the suspect's expected structure. Combine these four scores via weighted ensemble (default: weighted sum) to produce a final feature–suspect proximity score. Rank all candidates by this score and apply a threshold filter to generate a prioritized compound list for downstream validation.

Related tools

• Scannotation (Automated suspect screening tool that implements the MS1 predictor combiner scoring pipeline to compute feature–suspect proximity scores and prioritize compound candidates for LC-HRMS pre-annotation.) — https://github.com/scannotation/Scannotation_software

Evaluation signals

• Proximity scores are computed for all feature–suspect pairs without missing values; check that output table has dimensions = (n_features × n_suspects).
• Component scores (m/z, RT, isotope, neutral loss) fall within their expected ranges (typically 0–1 or 0–100 depending on normalization); verify no NaN, Inf, or out-of-bound values.
• Ranked output is sorted in descending order by proximity score; spot-check that top-ranked candidates have higher scores than lower-ranked ones.
• Filtered output respects the applied threshold: all remaining feature–suspect pairs have proximity score ≥ threshold; all removed pairs have score < threshold.
• For known positive control feature–suspect pairs, verify that the assigned proximity score is among the top-ranked (e.g., top 5%) for that feature, confirming the scoring mechanism is not inverting or randomizing ranks.

Limitations

• Scoring accuracy depends critically on the quality and completeness of reference data in the suspect compound database (isotope ratios, neutral loss fragments). Missing or incorrect reference patterns will degrade ranking.
• Retention time prediction quality varies across chromatographic methods and column chemistries; RT-based scoring may be unreliable if predicted RT windows are poorly calibrated or method-specific.
• The weighted ensemble method for combining component scores is heuristic; optimal weights may vary by chemical class or sample matrix, requiring method-specific tuning.
• Scannotation was developed and tested on Windows 10 and Mac (in Windows virtual machine); cross-platform compatibility on native Linux or other OS is not documented.

Evidence

• [readme] Scannotation combines MS1 chemical predictors and scores proximity for feature-suspect prioritization.: "This software combines several MS1 chemical predictors: m/z, retention times, isotopic patterns and neutral loss patterns, to score the proximity between features and suspects, thus efficiently"
• [other] The workflow involves loading data, computing four component scores, and combining them via ensemble method.: "Load feature table (m/z values, retention times, isotopic pattern signatures) and suspect compound database (reference m/z, expected retention time windows, isotope ratios, neutral loss fragments)."
• [readme] Scannotation enables rapid pre-annotation of LC-HRMS datasets through automated suspect screening.: "Scannotation is an automated and user-friendly suspect screening tool for the rapid pre-annotation of LC-HRMS datasets."
• [readme] Development and testing platform details.: "Scannotation was developed on Windows 10 and tested on both Windows and on a Mac computer in a Windows virtual machine."