Scalar coupling constants are a measurement of the nuclear spin interaction between pairs of atoms. They can be observed experimentally by nuclear magnetic resonance (NMR) spectroscopy. The measured values of scalar coupling constants can be directly related to molecular geometry and therefore can be used to derive detailed conformational information of proteins and other molecules, which is of broad importance in drug discovery and materials science. The extensive range of applications for scalar couplings make their prediction with machine learning a desirable objective.

This webinar will briefly introduce scalar couplings and then cover the successful methods utilized by the NVIDIA team during the Kaggle competition to predict molecular scalar couplings. We will demonstrate how NVIDIA libraries like RAPIDS can accelerate the development process.

• What a scalar coupling constant is and how it is utilized by various scientific fields

• Key techniques for predicting this property from molecular structure data, including key customizations and improvements on a Message Passing Neural Network (MPNN)

• End-to-end GPU examples demonstrating how NVIDIA libraries such as RAPIDS can accelerate this process with ~10 – 100X speedup

• Feature engineering and extraction techniques for tabular and graph data using cuDF

Content

content goes here