---
license: cc-by-nc-4.0
dataset_info:
  features:
  - name: graph_data
    struct:
    - name: edge_attr
      sequence:
        sequence: int64
    - name: edge_index
      sequence:
        sequence: int64
    - name: pos
      sequence:
        sequence: float64
    - name: solvent_class
      dtype: int64
    - name: x
      sequence:
        sequence: float64
  - name: c_peaks
    sequence:
      sequence: float64
  - name: h_peaks
    sequence:
      sequence: float64
  - name: filename
    dtype: string
  splits:
  - name: train
    num_bytes: 176931469
    num_examples: 21869
  - name: eval
    num_bytes: 3856453
    num_examples: 479
  - name: eval_zeroshot
    num_bytes: 1528168
    num_examples: 170
  - name: eval_fewshot
    num_bytes: 1636520
    num_examples: 188
  download_size: 44790710
  dataset_size: 183952610
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: eval
    path: data/eval-*
  - split: eval_zeroshot
    path: data/eval_zeroshot-*
  - split: eval_fewshot
    path: data/eval_fewshot-*
---

# 2DNMRGym

<!-- Provide a quick summary of the dataset. -->

This dataset contains more than 22,000 2D HSQC spectrum for atom-level graph learning tasks. 

To use this data, **please cite:** [https://arxiv.org/abs/2505.18181]

## Dataset Details

### Dataset Description

<!-- Provide a longer summary of what this dataset is. -->
HSQC is one important type of 2D NMR experiments, which is crucial in molecular discrimination and structure elucidation tasks. In 2DNMRGym, the train split is annotated by an established algorithm, which provides silver-standard labels for surrogated training. The evluation split is annotated by human experts with more than 10 years of training in organic chemistry, and provids high-quality gold-standard labels for model testing. 

### Dataset Sources [optional]

<!-- Provide the basic links for the dataset. -->
The raw data is sourced from HMDB and CH-NMR-NP.

- **Repository:** [https://github.com/siriusxiao62/Mol2DNMR]


## Dataset Structure
<!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. -->

Graph: dictionary presentation of pyg graph data. This is the molecular graph for each spectrum, containing node features and 3D coordinants. 

C-peaks: the carbon chemical shift of the cross peaks in each HSQC spectrum, annotated to match the carbon indices in the molecular graph. 

H-peaks: the proton chemical shifts of the cross peaks in each HSQC spectrum, annotated using the corresponding carbon indicies.


## Who are the annotators?

<!-- This section describes the people or systems who created the annotations. -->

Dr. Hao Xu - Harvard Medical School.

Dr. Wang Duosheng - Boston College.

Dr. Kumar Ambrish - University of Georgia.