Spherical Message Passing for 3D Molecular Graphs
Authors: Yi Liu, Limei Wang, Meng Liu, Yuchao Lin, Xuan Zhang, Bora Oztekin, Shuiwang Ji
ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate that the use of meaningful 3D information in Sphere Net leads to significant performance improvements in prediction tasks. Our results also demonstrate the advantages of Sphere Net in terms of capability, efficiency, and scalability. Our code is publicly available as part of the DIG library (https://github.com/divelab/DIG). |
| Researcher Affiliation | Academia | Yi Liu , Limei Wang , Meng Liu, Yuchao Lin, Xuan Zhang, Bora Oztekin & Shuiwang Ji Department of Computer Science & Engineering Texas A&M University College Station, TX 77843, USA {yiliu,limei,mengliu,kruskallin,xuan.zhang,bora,sji}@tamu.edu |
| Pseudocode | No | The paper includes figures illustrating the architecture and message passing scheme, along with mathematical equations, but it does not contain a dedicated pseudocode block or algorithm description. |
| Open Source Code | Yes | Our code is publicly available as part of the DIG library (https://github.com/divelab/DIG). Code is integrated in the DIG library (Liu et al., 2021) and available at https://github.com/divelab/DIG. |
| Open Datasets | Yes | We apply our Sphere Net to three benchmark datasets, including Open Catalyst 2020 (OC20) (Chanussot et al., 2020), QM9 (Ramakrishnan et al., 2014), and MD17 (Chmiela et al., 2017). |
| Dataset Splits | Yes | The dataset is original split into three sets, where the training set contains 110,000, the validation set contains 10,000, and the test set contains 10,831 molecules. (for QM9). Performance is evaluated on the validation set, which has four splits including In Domain (ID), Out of Domain Adsorbates (OOD Ads), Out of Domain catalysts (OOD cat), and Out of Domain Adsorbates and catalysts (OOD Both), where numbers of structures are 24,943, 24,961, 24,963, 24,987, respectively. (for OC20). |
| Hardware Specification | Yes | For QM9 and MD17 datasets, all models are trained using one NVIDIA Ge Force RTX 2080 Ti 11GB GPU. For the OC20 dataset, all models are trained using four NVIDIA RTX A6000 48GB GPUs. |
| Software Dependencies | No | Pytorch is used to implement all methods. However, specific version numbers for PyTorch or other software dependencies are not provided. |
| Experiment Setup | Yes | Detailed experimental setup is provided in Appendix D. Implementation hyper-parameters of Sphere Net on all the three datasets OC20, QM9, and MD17 are given in Table 6, Table 7, and Table 8, respectively. |