Molecular Weight Prediction From Ei Ms

molecular-weight-prediction-from-ei-ms

Summary

Use the MWFormer transformer model to directly predict molecular mass from electron ionization mass spectrometry (EI-MS) data by loading a pre-trained model and running forward inference on parsed m/z and intensity spectrum values. This skill enables rapid, direct mass inference without manual spectral interpretation or fragmentation pattern analysis.

When to use

Apply this skill when you have EI-MS spectrum data (m/z and intensity pairs) from a sample and need a direct, model-based prediction of molecular weight. This is most appropriate when you want to leverage transformer-based learned patterns over EI-MS instead of manual spectral interpretation, and when you have access to a pre-trained MWFormer model checkpoint.

When NOT to use

• Input spectrum is from a non-EI ionization source (e.g. ESI, APCI, MALDI) — MWFormer is trained specifically on electron ionization mass spectra.
• Input spectrum is already a processed feature vector or embedding — this skill expects raw or normalized m/z and intensity pairs.
• You have no access to a pre-trained MWFormer model and cannot train one — the skill assumes a trained model is available.

Inputs

• EI-MS spectrum data (m/z and intensity value pairs)
• Pre-trained MWFormer model checkpoint (.pt or .pth file)
• Spectrum configuration or normalization parameters matching model training

Outputs

• Predicted molecular weight (scalar mass value)
• Inference confidence or probability (if model outputs uncertainty)

How to apply

Set up a conda environment with Python 3.9, PyTorch 1.12 with CUDA 11.8 support, and rdkit. Clone the MWFormer repository from GitHub and install dependencies from requirements.txt. Load the pre-trained MWFormer model checkpoint from the repository. Parse your input EI-MS spectrum as m/z and intensity values, then prepare it as a tensor matching the model's expected input format (typically normalized spectrum data). Run forward inference through the MWFormer transformer to generate a direct molecular weight prediction. The output is a single predicted molecular mass value; success is indicated by a numerically valid mass prediction and absence of NaN or out-of-range outputs.

Related tools

• MWFormer (Pre-trained transformer neural network model that accepts EI-MS spectrum tensors and outputs molecular weight predictions) — https://github.com/zhanghailiangcsu/MWFormer
• PyTorch (Deep learning framework used to load the model and execute forward inference) — https://pytorch.org/
• Python (Programming language for environment setup, data parsing, and inference orchestration)
• Anaconda (Environment and dependency manager for Python 3.9 and conda packages (rdkit)) — https://www.anaconda.com
• rdkit (Cheminformatics toolkit (optional utility for validation or post-processing of predicted molecular weights))

Evaluation signals

• Predicted molecular weight is a positive real number within chemically plausible range (e.g. 10–2000 Da for most organic molecules).
• Forward inference completes without NaN, infinity, or runtime errors on valid EI-MS input tensors.
• Spectrum tensor shape and dtype match model input specification (e.g. batch, m/z bins, intensity values).
• Output molecular weight is consistent across multiple runs with identical input (deterministic given fixed model and input).
• If ground-truth molecular weights are available, compare predicted values to known masses; check for systematic bias or outliers beyond expected model error margins.

Limitations

• MWFormer is trained specifically on EI-MS data; it is not expected to generalize to spectra from other ionization methods (ESI, MALDI, APCI, etc.).
• Model performance depends on spectrum quality and preprocessing (normalization, m/z alignment) matching the training protocol — deviations may degrade accuracy.
• No changelog is provided in the repository, making version history and model updates opaque.
• Prediction is a single mass value; the model does not report confidence intervals, uncertainty estimates, or alternative candidates.
• The README does not specify input spectrum dimensions, preprocessing requirements, or acceptable m/z and intensity ranges; users must infer these from example code or model source.

Evidence

• [readme] MWFormer: Direct Prediction of Molecular Mass from Electron lonization Mass Spectrum by Transformer: "MWFormer: Direct Prediction of Molecular Mass from Electron lonization Mass Spectrum by Transformer"
• [other] The MWFormer model accepts electron ionization mass spectrum data as input and outputs a direct prediction of the molecular weight through a forward inference operation.: "The MWFormer model accepts electron ionization mass spectrum data as input and outputs a direct prediction of the molecular weight through a forward inference operation."
• [other] Parse the input EI-MS spectrum data (m/z and intensity values). Prepare the spectrum as a tensor input matching the model's expected input format. Run forward inference through the MWFormer transformer to generate the predicted molecular weight.: "Parse the input EI-MS spectrum data (m/z and intensity values). Prepare the spectrum as a tensor input matching the model's expected input format. Run forward inference through the MWFormer"
• [readme] conda create -n MWFormer python=3.9 conda activate MWFormer git clone https://github.com/zhanghailiangcsu/MWFormer.git pip install -r requirements.txt pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118 conda install -c conda-forge rdkit: "conda create -n MWFormer python=3.9 conda activate MWFormer git clone https://github.com/zhanghailiangcsu/MWFormer.git pip install -r requirements.txt pip3 install torch torchvision torchaudio"
• [readme] You can download the trained model on Github. Then refer to the example to use the model for prediction, and directly obtain the molecular weight from EI-MS.: "You can download the trained model on Github. Then refer to the example to use the model for prediction, and directly obtain the molecular weight from EI-MS."