Notice: This website is continuously updated as new conference proceedings and journal volumes are published. Consequently, the figures presented here may differ from those reported in our publications.
A brief overview of our methodology is provided on the Methods page, along with references to publications documenting our approach and links to the code and data used in each study.
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.
GraphDF: A Discrete Flow Model for Molecular Graph Generation
Authors: Youzhi Luo, Keqiang Yan, Shuiwang Ji
ICML 2021 | Venue PDF | LLM Run Details | Input Tokens: 20,529 Total number of tokens sent to the LLM as input for this paper's analysis. | Output Tokens: 4,938 Total number of tokens produced by the LLM (including reasoning/thinking tokens) for this paper's analysis.
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Comprehensive experimental results show that Graph DF outperforms prior methods on random generation, property optimization, and constrained optimization tasks. In this section, we evaluate our Graph DF on three tasks of molecule generation, as mentioned in Section 2.1. |
| Researcher Affiliation | Academia | 1Department of Computer Science & Engineering, Texas A&M University, TX, USA. |
| Pseudocode | No | The paper states, 'We summarize the detailed generation and training algorithms of Graph DF in Appendix A and B, respectively.', but these appendices containing pseudocode or algorithm blocks are not provided in the main paper text. |
| Open Source Code | Yes | The implementation of Graph DF has been integrated into the DIG (Liu et al., 2021) framework. |
| Open Datasets | Yes | For random generation of molecular graphs, we train and evaluate Graph DF on three molecule datasets, ZINC250K (Irwin et al., 2012), QM9 (Ramakrishnan et al., 2014), and MOSES (Polykovskiy et al., 2020). |
| Dataset Splits | No | The paper mentions training on datasets like ZINC250K and evaluating on generated graphs, but does not explicitly provide details about training, validation, or test dataset splits (e.g., percentages or specific partitioning methods). |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used for running experiments. |
| Software Dependencies | No | The paper mentions using 'Adam optimizer' and 'RDKit (Landrum, 2016)', but does not provide specific version numbers for general software dependencies or libraries required for reproducibility. |
| Experiment Setup | Yes | The flow model in Graph DF consists of 12 modulo shift modules, where the shared R-GCN has 3 message passing layers. The hidden dimension and output dimension of node embeddings are 128. ... Adam (Kingma & Ba, 2015) optimizer is used to train Graph DF models. On each molecule dataset, Graph DF is trained for 10 epochs, where the batch size is 32 and the learning rate is 0.001. ... where we fix the temperature τ = 0.1 in experiments. |