AI helps in Drug Discovery

 

Drug development for the treatment of various diseases is the cornerstone of medicine for human health. Drug development is an expensive and time-consuming process. Most drugs developed up to the human-trials stage end up having no effect at all or too many side-effects. These drugs are thus discarded resulting in wasted time, money and effort. The cost of these drugs end up on the price tags of drugs that do become successful.

Before drug development can even take place the drug discovery process must occur. Briefly, drug discovery involves a process of finding promising drug-like molecules that can bind or "dock" properly onto certain protein targets. After successfully docking to the protein, the binding drug, also known as the ligand, can stop a protein from functioning. If this happens to an essential protein of a bacterium, it can kill the bacterium, conferring protection to the human body.

Drug discovery, the process of discovering new candidate medications, involves combing through a mind-boggling number of potentially useful molecules. The number is estimated to be a novemdecillion, or 10^60 molecules! To put it into perspective, the Milky Way has about 100 thousand million, or 10^8 stars.

Here's where AI comes to the rescue. Researchers at the Massachusetts Institute of Technology have developed a new geometric deep learning model called EquiBind that is 1,200 times faster than one of the fastest existing computational molecular docking models. EquiBind uses the idea of "blind docking" where the model directly predicts the docking location without having to fitness score each docking one-by-one. Unlike most models that require several attempts to find a favorable position for the ligand in the protein, EquiBind already has built-in geometric reasoning that helps the model learn the underlying physics of molecules and successfully generalize to make better predictions when encountering new, unseen data.

The work is still ongoing in this area and the researchers have received feedback from industry to explore further applications of this technique.

The research paper for this advancement can be found here.