New research that combines machine learning techniques and material science could lead to breakthroughs in making drugs more effective or materials easier to degrade.
Researchers at the Machine Learning Center at Georgia Tech (ML@GT), Le Song, associate professor in the School of Computational Science and Engineering, and Rampi Ramprasad, professor in the School of Materials Sciences and Engineering), have recently been granted funding from the National Science Foundation (NSF) to further explore these possibilities.
The grant, worth $449,999 over three years, will explore using machine learning algorithms and material science techniques to study the molecules of different materials.
Molecules have long been “mapped” by assigning them unique fingerprints, and Song and Ramprasad are working toward making the fingerprints adaptive.
“By creating an adaptive fingerprint that can accurately predict a material’s properties, we are then able to search for a good material design much faster than with past methods,” said Song, associate director of ML@GT.
The fingerprints will encode the atomic arrangement of a molecule which can then be mapped by a graph neural network. By mapping the fingerprints, scientists are able to take into account the symmetry that exists in materials and can design better deep learning models to further improve their work.
The duo’s research will focus on a particular class of material system including molecules, polymers, and materials made out of carbon, hydrogen, oxygen, and nitrogen.