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.
main_flowchart
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}
}
