Python
Download and load TrialBench datasets in Python.
Quick Install
pip install trialbenchTrialBench
Build and benchmark AI for clinical trial prediction with 23 ready-to-use datasets spanning 8 tasks. TrialBench brings together molecular, disease, text, and tabular features, with Python and R access to simplify data preparation and model evaluation.
6,538 data & package downloads
As of 2 October 2026.
R
Use TrialBench datasets in your R workflow.
If you find this resource useful, please cite:
Scientific Data 2025 · BibTeX
@article{chen2025trialbench,
title = {{TrialBench}: Multi-Modal AI-Ready Datasets for Clinical Trial Prediction},
author = {Chen, Jintai and Hu, Yaojun and Cai, Mingchen and Lu, Yingzhou and Wang, Yue and Cao, Xu and Lin, Miao and Xu, Hongxia and Wu, Jian and Cao, Xiao and Sun, Jimeng and Li, Yuqiang and Glass, Lucas and Huang, Kexin and Zitnik, Marinka and Fu, Tianfan},
journal = {Scientific Data},
volume = {12},
pages = {1564},
year = {2025},
doi = {10.1038/s41597-025-05680-8}
}Electrocardio Panorama
Resources for reconstructing ECG signals beyond fixed lead configurations, from waveform annotations to dense recordings with spatial coordinates.
ECG Waveform Annotation Toolkit
Waveform annotations and an interactive labeling tool help researchers locate P waves, QRS complexes, and T waves, segment heartbeats, and train ECG delineation models. Standard 12-lead viewing angles also support research on spatially conditioned ECG synthesis.
If you find this resource useful, please cite:
IJCAI 2021 · BibTeX
@inproceedings{chen2021electrocardio,
title = {Electrocardio Panorama: Synthesizing New {ECG} Views with Self-supervision},
author = {Chen, Jintai and Zheng, Xiangshang and Yu, Hongyun and Chen, Danny Z. and Wu, Jian},
booktitle = {Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence (IJCAI-21)},
pages = {3597--3605},
year = {2021},
doi = {10.24963/ijcai.2021/495}
}Panobench: 48-View ECG Dataset
A 48-view ECG dataset with 4,470 ten-second recordings and CT-derived viewing angles. Its 6 limb and 42 precordial leads enable researchers to reconstruct dense ECG views from sparse inputs and evaluate new-view synthesis against recorded signals, beyond the coverage of standard 12-lead ECG.
If you find this resource useful, please cite:
ICLR 2026 · BibTeX
@inproceedings{zhan2026nefnet,
title = {{Nef-Net v2}: Adapting Electrocardio Panorama in the wild},
author = {Zhan, Zehui and Hu, Yaojun and Zhang, Jiajing and Lian, Wanchen and Wu, Wanqing and Chen, Jintai},
booktitle = {The Fourteenth International Conference on Learning Representations},
year = {2026},
url = {https://openreview.net/forum?id=JzZhhhxniR}
}