MIT Researchers have developed a geometric depth learning model . The model is successful in binding drug like molecules to proteins , Faster than the fastest computational molecular docking model 、 More accurate , It reduces the opportunity and cost of drug test failure .
as everyone knows , The whole universe is full of countless molecules .
How many of these molecules have potential drug like properties , It can be used to develop life-saving drugs ？ It's a million ？ Or a billion ？ Or trillions ？
The answer is ：10 Of 60 The next power .
Such a huge number , It has greatly delayed the research and development progress of new drugs , Fast spreading diseases like covid-19 , At present, there is no specific drug , It's also because the types and numbers of molecules are too large , It is far beyond the scope of existing drug design models .
MIT A research team in does not believe this evil . It doesn't count, does it , It's OK to add acceleration to the previous model ？
This acceleration , Namely 1200 times .
They studied a product called 「EquiBind」 Geometry depth learning model , This model is faster than the previous fastest computational molecular docking model 「QuickVina2-W」 fast 1200 times , Successfully bind drug like molecules to proteins , It reduces the opportunity and cost of drug test failure .
The research paper will be published in ICML 2022 On .
