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..
Atomic Diffusion Models for Small Molecule Structure Elucidation from NMR Spectra
Authors: Ziyu Xiong, Yichi Zhang, Foyez Alauddin, Chu Xin Cheng, Joon An, Mohammad Seyedsayamdost, Ellen D. Zhong
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We compare CHEFNMR against chemical language model-based and graph-based formulations and demonstrate state-of-the-art accuracy across multiple synthetic and experimental benchmarks. |
| Researcher Affiliation | Academia | Ziyu Xiong Princeton University EMAIL Yichi Zhang Princeton University EMAIL Foyez Alauddin Princeton University EMAIL Chu Xin Cheng California Institute of Technology EMAIL Joon Soo An Princeton University EMAIL Mohammad R. Seyedsayamdost Princeton University EMAIL Ellen D. Zhong Princeton University EMAIL |
| Pseudocode | Yes | Algorithm 1 Diffusion Training. Algorithm 2 Smooth LDDT Loss. Algorithm 3 Diffusion Sampling using Stochastic Heun s 2nd order Method. |
| Open Source Code | No | We will release our datasets (See details in Appendix C). We plan to release our model upon publication of the method. |
| Open Datasets | Yes | We evaluate models on two public benchmarks, Spectra Base [22] and USPTO [4], and our self-curated Spectra NP dataset. ... The original dataset is available at https://zenodo.org/records/13892026 under the CC-BY 4.0 license. ... The original dataset is available at https://zenodo.org/records/11611178 under the Community Data License Agreement-Sharing 1.0 (CDLA-Sharing-1.0). ... Spec Teach dataset [65] ... The original dataset is available at https://drive.google.com/ drive/folders/1R23KGk3bp6uk GCRb4U-CRuxn L6PYYBYc under CC-BY 4.0. ... NMRShift DB2 [34] ... The original dataset is under the nmrshiftdb2 Database License (https://nmrshiftdb.nmr.uni-koeln.de/nmrshiftdbhtml/ nmrshiftdb2datalicense.txt). |
| Dataset Splits | Yes | The original dataset comprises 142,894 tuples of (Canonical nonstereo SMILES, 28,000-dimensional 1H NMR spectrum, 80-bin 13C NMR spectrum) along with non-overlapping split indices in a ratio of 0.8:0.1:0.1 for training, validation, and test sets. ... The original split indices are preserved, resulting in a post-filtering split ratio of 0.86:0.04:0.1 for training, validation, and test sets. ... The dataset is randomly split into training, validation, and test sets in a ratio of 0.8:0.1:0.1. |
| Hardware Specification | Yes | Evaluation is conducted on 1 A100 GPU, with runtime varying between 30 minutes and 2 hours depending on the dataset. Other hyperparameters are set as default in the original model [22]. ... NMR-Di Gress ... is trained on each dataset using 4 A100 GPUs for 48 hours. ... Sampling time is the estimated average time on 1 A100 or H100 GPU for three independent runs. |
| Software Dependencies | No | The paper mentions RDKit and Mestre Nova but does not provide specific version numbers for these or any other software libraries or programming languages used. |
| Experiment Setup | Yes | Appendix Table 8: CHEFNMR hyperparameters and optimizer settings. Parameter CHEFNMR-S CHEFNMR-L NMR-Conv Former General Encoder Dimension (Dencoder) 256 Dropout Rate 0.1 ... Optimizer (Adam) Learning Rate 1e-4 Adam β1 0.9 Adam β2 0.95 Adam ϵ 1e-8 |