MolBench: A Benchmark of AI Models for Molecular Property Prediction

• First Online: 14 February 2024

Abstract

In recent years, there has been a growing demand for the prediction of complex molecular properties in the fields of drug design, material science, and biotechnology. Compared to traditional laboratory methods, the deep learning method has many advantages such as saving enormously time and money. The deep learning method achieves revolutionary success in predicting molecular properties and many models based on the deep learning method has been developed in this field. However, there still lacks reliable and multidimensional benchmarks for evaluating these artificial intelligence (AI) models. In this paper, we develop a general method to evaluate AI models for predicting molecular properties. More precisely, we design multiple evaluation metrics based on the MoleculeNet datasets and introduce an extensible API interface to benchmark three types of AI models: molecular fingerprint based models, graph-based models, and pre-trained models. The purpose of the work is to establish a fair and reliable benchmark for future innovation in the field of molecular property prediction, emphasizing the importance of multidimensional perspectives.

Download paper

Recommended Citation

• Jiang X, Tan L, Cen J, et al. MolBench: A Benchmark of AI Models for Molecular Property Prediction[C] International Symposium on Benchmarking, Measuring and Optimization. Singapore: Springer Nature Singapore, 2023: 53-70.

BibTeX

@inproceedings{jiang2023molbench,
  title={MolBench: A Benchmark of AI Models for Molecular Property Prediction},
  author={Jiang, Xiuyu and Tan, Liqin and Cen, Jianhuan and Zou, Qingsong},
  booktitle={International Symposium on Benchmarking, Measuring and Optimization},
  pages={53--70},
  year={2023},
  organization={Springer}
}