This is the code for the Paper: DiffSpectra: Molecular Structure Elucidation from Spectra using Diffusion Models.
DiffSpectra is a generative framework for molecular structure elucidation from multi-modal spectral data. Unlike retrieval-based approaches that rely on finite molecular libraries or SMILES-based autoregressive models that often ignore 3D geometry, DiffSpectra formulates structure elucidation as a conditional diffusion process.
The framework integrates two core components:
- Diffusion Molecule Transformer (DMT): An SE(3)-equivariant denoising network that models both 2D topology and 3D geometry of molecules.
- SpecFormer: A transformer-based spectral encoder that captures intra- and inter-spectrum dependencies across diverse spectral modalities (e.g., IR, Raman, UV-Vis).
Through spectrum-conditioned diffusion modeling, DiffSpectra unifies multi-modal reasoning with 2D/3D generative modeling. Extensive experiments demonstrate that DiffSpectra achieves 40.76% top-1 accuracy and 99.49% top-10 accuracy in recovering exact molecular structures.
To our knowledge, DiffSpectra is the first framework that unifies multi-modal spectral reasoning and joint 2D/3D generative modeling for de novo molecular structure elucidation.
- Paper: https://arxiv.org/abs/2507.06853
- Code: https://github.com/AzureLeon1/DiffSpectra
- Dataset: https://huggingface.co/datasets/AzureLeon1/DiffSpectra
- Model Checkpoins: https://huggingface.co/AzureLeon1/DiffSpectra
- Install the environment according to the commands in
env.sh. - Download and place the dataset in your configured
path/to/datasetdirectory. - Modify the configuration file in
configs/to setdata.rootto your configured path. - [Optional] Download and place the trained model checkpoints in the
exp/directory. - Use the provided scripts in
scripts/for training or evaluation.
You can train the model with different spectra settings as follows:
-
All spectra with pretrained SpecFormer
CUDA_VISIBLE_DEVICES=0,1 python main.py \ --config configs/diffspectra_qm9s.py \ --config_original_qm9 configs/base_qm9.py \ --mode train \ --workdir exp/allspectra_pretrained_specformer \ --config.data.spectra_version allspectra \ --config.model.name DMT \ --config.model.pretrained_specformer_path exp/pretrained_specformer.ckpt
-
All spectra
CUDA_VISIBLE_DEVICES=0,1 python main.py \ --config configs/diffspectra_qm9s.py \ --config_original_qm9 configs/base_qm9.py \ --mode train \ --workdir exp/allspectra \ --config.data.spectra_version allspectra \ --config.model.name DMT
-
Only IR
CUDA_VISIBLE_DEVICES=0,1 python main.py \ --config configs/diffspectra_qm9s.py \ --config_original_qm9 configs/base_qm9.py \ --mode train \ --workdir exp/ir \ --config.data.spectra_version ir \ --config.model.name DMT
-
Only Raman
CUDA_VISIBLE_DEVICES=0,1 python main.py \ --config configs/diffspectra_qm9s.py \ --config_original_qm9 configs/base_qm9.py \ --mode train \ --workdir exp/raman \ --config.data.spectra_version raman \ --config.model.name DMT
-
Only UV-Vis
CUDA_VISIBLE_DEVICES=0,1 python main.py \ --config configs/diffspectra_qm9s.py \ --config_original_qm9 configs/base_qm9.py \ --mode train \ --workdir exp/uv \ --config.data.spectra_version uv \ --config.model.name DMT
To evaluate the trained model, first generate molecules according to the spectra:
-
All spectra with pretrained SpecFormer
CUDA_VISIBLE_DEVICES=0,1 python main.py \ --config configs/diffspectra_qm9s.py \ --config_original_qm9 configs/base_qm9.py \ --mode eval \ --workdir exp/allspectra_pretrained_specformer \ --config.eval.ckpts '40' \ --config.eval.num_samples 10000 \ --config.eval.save_mols true \ --config.data.spectra_version allspectra
-
All spectra
CUDA_VISIBLE_DEVICES=0,1 python main.py \ --config configs/diffspectra_qm9s.py \ --config_original_qm9 configs/base_qm9.py \ --mode eval \ --workdir exp/allspectra \ --config.eval.ckpts '40' \ --config.eval.num_samples 10000 \ --config.eval.save_mols true \ --config.data.spectra_version allspectra
-
Only IR
CUDA_VISIBLE_DEVICES=0,1 python main.py \ --config configs/diffspectra_qm9s.py \ --config_original_qm9 configs/base_qm9.py \ --mode eval \ --workdir exp/ir \ --config.eval.ckpts '40' \ --config.eval.num_samples 10000 \ --config.eval.save_mols true \ --config.data.spectra_version ir
-
Only Raman
CUDA_VISIBLE_DEVICES=0,1 python main.py \ --config configs/diffspectra_qm9s.py \ --config_original_qm9 configs/base_qm9.py \ --mode eval \ --workdir exp/raman \ --config.eval.ckpts '40' \ --config.eval.num_samples 10000 \ --config.eval.save_mols true \ --config.data.spectra_version raman
-
Only UV-Vis
CUDA_VISIBLE_DEVICES=0,1 python main.py \ --config configs/diffspectra_qm9s.py \ --config_original_qm9 configs/base_qm9.py \ --mode eval \ --workdir exp/uv \ --config.eval.ckpts '40' \ --config.eval.num_samples 10000 \ --config.eval.save_mols true \ --config.data.spectra_version uv
After generating molecules, compute metrics based on the generated molecules. For accuracy and similarity evaluation, these metrics will be computed automatically along with the generation process. Alternatively, you can also compute them manually by running:
python eval_sampled_mols/base_metrics.pyFor 3D structural evaluation, you can run:
python eval_sampled_mols/rmsd.pyThe generated molecules and evaluation metrics will be found at exp/{exp_name}/eval/.
For questions, feedback, or collaborations, please reach out to: 📧 liang.wang@cripac.ia.ac.cn
If you use DiffSpectra in your research or applications, please cite:
@article{wang2025diffspectra,
title={DiffSpectra: Molecular Structure Elucidation from Spectra using Diffusion Models},
author={Liang Wang and Yu Rong and Tingyang Xu and Zhenyi Zhong and Zhiyuan Liu and Pengju Wang and Deli Zhao and Qiang Liu and Shu Wu and Liang Wang and Yang Zhang},
journal={arXiv},
volume={abs/2502.09511},
year={2025}
}