Suiren-ConfAvg

Suiren-ConfAvg is derived from the Suiren-Base model through distillation, designed to characterize the conformational average representations of molecules. In short, Suiren-Base can provide microscopic representations of various molecular conformations, but many scientific tasks rely on the ensemble averaging of multiple conformations. We compressed the features of Suiren-Base into Suiren-ConfAvg through a special distillation method, whose representations can be used to solve some macroscopic tasks, such as property prediction and molecular generation. We provide a quick training script for property prediction on GitHub.

Usage

You can use the scripts we provide in github to train your data directly. Alternatively, if you want to use the Suiren-ConfAvg model directly, you can call the GNN class in the file. The configuration for Suiren-ConfAvg is:

model = GNN(
            num_layer=12,
            emb_dim=256,
            drop_ratio=0.0,
            model_mode="pretrain",
            output_type="last" # "last" means returning the node embedding of the last layer (torch.tensor); "layers" means returning the node embedding of the all layers (list(torch.tensor))
        )
        

Citation

If you use Suiren models, please cite the relevant papers for the underlying models.

@article{an2026suiren,
  title={Suiren-1.0 Technical Report: A Family of Molecular Foundation Models},
  author={An, Junyi and Lu, Xinyu and Shi, Yun-Fei and Xu, Li-Cheng and Zhang, Nannan and Qu, Chao and Qi, Yuan and Cao, Fenglei},
  journal={arXiv preprint arXiv:2603.21942},
  year={2026}
}