In comparison to existing state-of-the-art techniques, researchers have created an artificial intelligence model that can locate potential therapeutic compounds more than 1,000 times faster.
The Massachusetts Institute of Technology (MIT) team claims that the EquiBind AI model will drastically lower the likelihood and expense of failed drug trials.
There are a staggering 1060 compounds that contain properties that might make them viable drugs. The Milky Way galaxy has around 108 stars, in contrast.
- Despite Ukraine war EU remains largest buyer of Russian oil
- A Democratic economic accord is complicated by Covid
- The S-300 missile didn’t harm the Bayraktar TB2!
One of the fastest computer molecular docking models available is 1,200 times slower than the EquiBind model’s ability to correctly bind these drug-like compounds to proteins.
This is made possible by EquiBind’s inherent geometric reasoning, which enables it to anticipate which proteins would fit to a molecule without knowing in advance where its target pocket is.
“We were amazed that while other methods got it completely wrong or only got one correct, EquiBind was able to put it into the correct pocket, so we were very happy to see the results for this,” said Hannes Stärk, a first-year graduate student at the MIT Department of Electrical Engineering and Computer Science and lead author of the paper describing the research.
Industry leaders have already expressed interest in the discoveries, which they think will help in the search for therapies for gastrointestinal tumors, leukemia, and lung cancer.
“EquiBind provides a unique solution to the docking problem that incorporates both pose prediction and binding site identification,” said Pat Walters, the chief data officer for drug discovery firm Relay Therapeutics.
“This approach, which leverages information from thousands of publicly available crystal structures, has the potential to impact the field in new ways.”
At the International Conference on Machine Learning, a work titled “EquiBind: Geometric Deep Learning for Drug Binding Structure Prediction” will be presented (ICML).
