If you click through to the blog post, it’s also clear that computational protein structure prediction basically “doesn’t work” currently, in the sense of being able to predict which drugs would bind to a protein, well enough to replace experiments. 40% accuracy? Ick.
-
-
Show this thread
-
Today’s machine learning isn’t enough to solve the problem, but it isn’t window dressing either. Accuracy in protein structure prediction has reliably improved as neural network architectures have gotten deeper and more powerful.
Show this thread -
A meta point: The practical question to ask of any ML algorithm is “what expensive manual process can this replace”? In biotech, that will usually mean “predict the result of an experiment so well you don’t have to run it” or “classify sample data so a human doesn’t have to”
Show this thread -
Initial binding-affinity experiments are automated and scalable enough today that any computational chemistry algorithm would have to pass a *really* high bar for it to usefully replace these experiments.
Show this thread
End of conversation
New conversation -
-
-
Nit really relevant, unfortunately.https://twitter.com/AndrewCutler13/status/1195053650353127425 …
-
You can usually get the structure from crystallography, and it usually doesn’t help.
- 2 more replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.