5 ms·
Geometric deep learning of RNA structure
- dekhn 5y agoThere's a lot of interesting computer science in the science of folding of RNA. For a very long time, parsing theory was the preferred way to do this, until people foudn examples of RNA that were derived from nonregular grammers (https://en.wikipedia.org/wiki/Pseudoknot https://en.wikipedia.org/wiki/Pseudoknot) making it hard to predict due to the high computational complexity. Much of the recent work has thrown nearly all the parsing theory away and instead uses learned geometric potentials to build scoring (ranking) functions that are much more accurate. Once again, "dumb" theory (deep neural nets) works better than "smart" theory (grammars, ab initio simulations).