MXMNet Implementation running on GPUs in nanoHUB.

This network makes prediction of properties from molecular structures. In this case, heat of formation. This is validated against the QM9 dataset of over 130,000 molecules. The architecure mimics the covalent and non-covalent interactions using a Graph Neural Network. 

This effort was supported by the US National Science Foundation (DMREF-1922316). We acknowledge computational resources from nanoHUB and Purdue University through the Network for Computational Nanotechnology.

This is an adaptation of the code from the following publication:

Zhang, S., Liu, Y., & Xie, L. (2020). Molecular mechanics-driven graph neural network with multiplex graph for molecular structures. arXiv preprint arXiv:2011.07457.

Researchers should cite this work as follows:

• Juan Carlos Verduzco Gastelum, Alejandro Strachan (2022), "GPU Implementation of MXMNet," https://nanohub.org/resources/mxmnetgpu. (DOI: 10.21981/QG2N-FB31).

  BibTex | EndNote

1. graph neural network
2. physics-based machine learning
3. property prediction