Overview
DiffMS generates candidate molecular structures conditioned on mass-spectral information using a diffusion model. The research pipeline combines fingerprint-to-molecule pretraining with end-to-end spectral fine-tuning and provides pretrained model resources.
What can this tool do?
Common uses
- DiffMS generates candidate molecular structures conditioned on mass-spectral information using a diffusion model.
Scientific fit
Primary function
Metabolite annotation and identification
Other functions
Machine learning and prediction
Data compatibility
Inputs
Mass spectra
Outputs
Molecular structures
Input formats
Prepared spectra, molecular formula and model configuration/checkpoints
Output formats
Generated molecular candidates and training/evaluation artifacts
Scope and considerations
Designed for
- DiffMS generates candidate molecular structures conditioned on mass-spectral information using a diffusion model.
Not designed for
- Confirmed molecular identification or a turnkey raw-vendor-file analysis pipeline.
Important requirements
- The documented Python/PyTorch/RDKit environment, prepared dataset configuration and appropriate end-to-end or separate encoder/decoder checkpoints; published setup uses CUDA. The published model is restricted by a supplied molecular formula.
Considerations
- Check data preparation, checkpoint compatibility and chemical-domain coverage. Preserve train/test separation and validate generated candidates with independent evidence.
Access and documentation
MIT source distribution. External datasets, dependent packages and any hosted routes retain their own terms; software operation was not tested.
License: MIT
Evidence and review
Enveda intake rechecked against official repository documentation, package/license metadata and primary publication identity on 2026-10-27. Editorial source review only; no software execution or independent benchmark. See docs/enveda-expansion-review-2026-10-27-batch3.md.
Core publications
Resource papers, direct updates and tool-specific methods. Selection policy.
- Bohde et al. DiffMS: Diffusion Generation of Molecules Conditioned on Mass Spectra. arXiv:2502.09571 (2025). Authors identify the associated conference as ICML 2025. doi:10.48550/arXiv.2502.09571
Related tools
Entry history
Changes to this catalogue entry, not the tool's software release history. Acceptance and editorial review are separate.
- 2026-10-27 · First committed entry
Add 45 source-reviewed Enveda tool entries in third expansion batch
View change · Entry at this revision