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..
MoleBridge: Synthetic Space Projecting with Discrete Markov Bridges
Authors: Rongchao Zhang, Yu Huang, Yongzhi Cao, Hanpin Wang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirically, we demonstrate that Mole Bridge excels in a variety of scenarios. 5 Experiments 5.1 Experimental Setup |
| Researcher Affiliation | Academia | 1Key Laboratory of High Confidence Software Technologies (Peking University), Ministry of Education, School of Computer Science, Peking University 2National Engineering Research Center for Software Engineering, Peking University EMAIL, EMAIL |
| Pseudocode | No | The paper describes the algorithm steps in paragraph form, such as in Section 4.1 'Sampling', but does not include a clearly labeled 'Pseudocode' or 'Algorithm' block. |
| Open Source Code | No | Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: The paper will open up data and code access when permissions allow, and provide instructions for doing so. |
| Open Datasets | Yes | Datasets. We use the Syn Net reaction template set [19] for reaction templates R, which is based on two publicly available template collections from Hartenfeller et al [24]. and Button et al [10]. For building blocks B, we use the Enamine US Stock catalog [1] as the data source. ... Additionally, we include a challenging test set: molecules extracted from the Ch EMBL database [20], which have been previously reported as unreachable target compounds [19, 46]. |
| Dataset Splits | Yes | We use K-means clustering based on Morgan fingerprints to group the blocks into 128 clusters, reserving one structurally distinctive cluster for testing and using the remaining 127 clusters for training. |
| Hardware Specification | Yes | The model is trained on 4 NVIDIA 4090 GPUs with a batch size of 128 and 4 data loader workers. |
| Software Dependencies | No | The paper mentions using RDKit [8] and the Adam optimizer [34], but does not provide specific version numbers for these or any other software dependencies. |
| Experiment Setup | Yes | The model is trained on 4 NVIDIA 4090 GPUs with a batch size of 128 and 4 data loader workers. We use the Adam optimizer [34] with an initial learning rate of 3 10 4, and momentum parameters β1 = 0.90, β2 = 0.999. A plateau-based learning rate scheduler is used, reducing the learning rate by a factor of 0.6 when validation performance plateaus, with patience of 5 validation cycles and a minimum learning rate of 1 10 5. |