|
|
--- |
|
|
license: apache-2.0 |
|
|
datasets: |
|
|
- AzureLeon1/DiffSpectra |
|
|
--- |
|
|
|
|
|
# DiffSpectra |
|
|
|
|
|
## Model Description |
|
|
**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](https://arxiv.org/abs/2507.06853) |
|
|
- **Code:** [https://github.com/AzureLeon1/DiffSpectra](https://github.com/AzureLeon1/DiffSpectra) |
|
|
- **Dataset:** [https://huggingface.co/datasets/AzureLeon1/DiffSpectra](https://huggingface.co/datasets/AzureLeon1/DiffSpectra) |
|
|
|
|
|
--- |
|
|
|
|
|
## Usage |
|
|
1. Place model checkpoints into the `exp/` directory. |
|
|
2. Use the provided scripts in `scripts/` for training or evaluation. |
|
|
|
|
|
Note: Please ensure required dependencies and environment settings match those described in the repository instructions. |
|
|
|
|
|
## Contact |
|
|
|
|
|
For questions, feedback, or collaborations, please reach out to: |
|
|
📧 [email protected] |
|
|
|
|
|
## Cite Us |
|
|
|
|
|
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} |
|
|
} |
|
|
``` |
