Ms ms spectrum annotation preprocessing

MS/MS Spectrum Annotation Preprocessing

Summary

Prepare and annotate MS/MS spectra from LC-MS/MS acquisitions in DDA mode by curating training datasets with purity labels to enable downstream chimeric spectrum purification and customized model training. This skill bridges raw spectral acquisition and machine-learning-ready training sets.

When to use

You have raw LC-MS/MS data acquired in Data-Dependent Acquisition (DDA) mode and need to create a labeled training dataset for customized purification model development. Specifically, when you want to train DNMS2Purifier or similar spectrum-quality classifiers on your own instrument, sample matrix, or experimental conditions rather than relying on pre-trained models.

When NOT to use

• Your input spectra are already pre-processed, quality-filtered, or de-chimeric from another tool — this skill assumes raw or minimally processed spectra requiring annotation.
• You are applying an existing pre-trained DNMS2Purifier model to new data without retraining — use the main DNMS2Purifier.r purification script directly instead.
• Your MS/MS data was acquired in targeted or Selected Reaction Monitoring (SRM) mode rather than Data-Dependent Acquisition (DDA) mode.

Inputs

• Raw MS/MS spectra from LC-MS/MS DDA acquisitions
• Annotated purity labels or quality scores for each spectrum
• Structured training data (data frame or matrix) containing spectral features and annotations

Outputs

• Serialized trained purification model (.rds file)
• Customized model object compatible with DNMS2Purifier.r workflow

How to apply

Collect MS/MS spectra from LC-MS/MS acquisitions performed in DDA mode, ensuring representative coverage of your experimental conditions and sample matrices. Manually or semi-automatically annotate each spectrum with a purity label (typically binary: pure or chimeric, or continuous quality scores) based on visual inspection, spectral quality metrics, or ground-truth tandem MS interpretation. Organize the annotated spectra and their purity labels into a structured dataset (e.g., a data frame with spectral features and labels). Load this training dataset into the DNMS2Purifier_model_generation.r script running under R 4.2.1, which will train a customized purification model on the annotated data. The trained model is serialized and saved as an .rds file for integration into the main DNMS2Purifier.r workflow.

Related tools

• DNMS2Purifier_model_generation.r (R script that trains the customized purification model on annotated training data and serializes the model to .rds format) — https://github.com/HuanLab/DNMS2Purifier
• R (Runtime environment (version 4.2.1 required) for executing model training and serialization)
• DNMS2Purifier.r (Main purification script that consumes the trained model .rds file for MS/MS spectrum purification) — https://github.com/HuanLab/DNMS2Purifier

Examples

# Load training data with annotated MS/MS spectra and purity labels, then execute:
Rscript DNMS2Purifier_model_generation.R --input training_data.csv --output customized_model.rds

Evaluation signals

• Training dataset contains annotated MS/MS spectra with corresponding purity labels; no missing or mismatched entries between spectra and labels.
• Trained model .rds file is successfully serialized and can be loaded into R 4.2.1 without errors.
• Model object contains expected structure and parameters (e.g., feature weights, classification thresholds) compatible with DNMS2Purifier.r intake.
• When the trained model is applied to held-out test spectra, classification accuracy, sensitivity, and specificity meet domain expectations for chimeric vs. pure spectrum discrimination.
• Training data distribution reflects the instrument, sample matrix, and ionization conditions of the target application (e.g., representative sample diversity, ion abundance ranges).

Limitations

• Annotation quality and completeness directly affect model performance; subjective or incomplete purity labeling will propagate as classification bias.
• Training dataset must be sufficiently large and representative to avoid overfitting; no minimum sample size is specified in the documentation.
• Model trained on one LC-MS/MS platform or sample matrix may not generalize well to different instruments, ion sources, or chemical spaces without retraining.
• No changelog or version history is publicly available, limiting reproducibility and tracking of model training updates.
• The skill is specific to the DNMS2Purifier framework; compatibility with other spectrum purification tools or formats is not addressed.

Evidence

• [other] Prepare training data containing annotated MS/MS spectra and purity labels from LC-MS/MS acquisitions in DDA mode.: "Prepare training data containing annotated MS/MS spectra and purity labels from LC-MS/MS acquisitions in DDA mode."
• [other] Execute the model generation script to train the customized purification model on the prepared training dataset.: "Execute the model generation script to train the customized purification model on the prepared training dataset."
• [other] Serialize and save the trained model object to an .rds file for downstream use in the DNMS2Purifier.r purification workflow.: "Serialize and save the trained model object to an .rds file for downstream use in the DNMS2Purifier.r purification workflow."
• [readme] The program is written in R (ver 4.2.1).: "The program is written in R (ver 4.2.1)."
• [readme] DNMS2Purifier is a bioinformatic solution that purifies chimeric MS/MS spectra from LC-MS/MS-based untargeted metabolomics in Data-Dependent Acquisition (DDA) mode.: "DNMS2Purifier is a bioinformatic solution that purifies chimeric MS/MS spectra from LC-MS/MS-based untargeted metabolomics in Data-Dependent Acquisition (DDA) mode."