Making virtual drug screening more efficient

Using a type of machine learning called graph convolutional networks, the artificial intelligence-powered model can simultaneously predict several properties of a drug, such as absorption, distribution, metabolism, excretion and toxicity.

Compared to existing methods, the new model is 100 times faster at screening drug molecules with minimal computational costs. It can also be run on a personal computer.

“Our all-in-one method removes the need for multiple specialised models to predict how a drug will function in the body, which could make the drug discovery process more efficient,” says Assoc Prof Mu.

Read more about the innovation in “Application of variational graph encoders as an effective generalist algorithm in computer-aided drug design”, published in Nature Machine Intelligence(2023), DOI: 10.1038/s42256-023-00683-9.

The article appeared first in NTU's research & innovation magazine Pushing Frontiers (issue #23, March 2024).