Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Molecule Joint Auto-Encoding: Trajectory Pretraining with 2D and 3D Diffusion
Authors: weitao Du, Jiujiu Chen, Xuecang Zhang, Zhi-Ming Ma, Shengchao Liu
NeurIPS 2023 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirically, Molecule JAE proves its effectiveness by reaching state-of-the-art performance on 15 out of 20 tasks by comparing it with 12 competitive baselines. |
| Researcher Affiliation | Collaboration | 1 Department of Mathematics, Chinese Academy of Sciences 2 Shenzhen Institute for Advanced Study, University of Electronic Science and Technology of China 3 Huawei Technologies Ltd 4 Department of Computer Science and Operations Research, Université de Montréal |
| Pseudocode | No | The paper describes its methodology through textual explanations and diagrams but does not include any formal pseudocode or algorithm blocks. |
| Open Source Code | No | The code is available on this website. (No URL is provided in the document.) |
| Open Datasets | Yes | For pretraining, we use PCQM4Mv2 [55]. and QM9 [59] is a dataset of 134K molecules... and MD17 [46] is a dataset on molecular dynamics simulation. |
| Dataset Splits | Yes | We take 110K for training, 10K for validation, and 11K for testing. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory, or processor types) used for running the experiments. |
| Software Dependencies | No | The paper mentions using specific models/architectures like Sch Net and Mi Di transformer, but it does not specify any software dependencies with version numbers (e.g., Python, PyTorch, CUDA versions). |
| Experiment Setup | Yes | Table 6: Hyperparameter specifications for Molecule JAE of property prediction. epochs {50, 100} learning rate {5e-4, 1e-4} β {[0.1, 10]} number of steps {1000} λ1 {1} λ2 {0, 0.01, 1} and Table 7: Hyperparameter specifications for Molecule JAE of molecule generation. epochs {3000, 10000} learning rate {2e-4, 3e-4} number of layers 12 number of diffusion steps {500} diffusion noise schedule cosine mlp hidden dimensions {X: 256, E: 128, y: 128, pos: 64} λtrain {X: 0.4, E: 2, y: 0, pos: 3, charges: 1} |