|
|
|
|
|
by dekhn
1603 days ago
|
|
I read that as a teenager, thought it sounded nice, went to grad school and did molecular dynamics simulations (like folding at home) for a decade, then went to google and built the world's largest simulation system (basically, the largest group of nodes running folding at home). Eventually we shut the system down because it was an inefficient way to predict protein structure and sample folding processes (although I got 3-4 excellent papers from it). The idea is great, it was a wonderful narrative to run my life for a while, but eventually, the more I learned, the more impractical using full atomistic simulations seem for solving any problem. It seems more likely we can train far more efficient networks that encapsulate all the salient rules of folding in a much smaller space, and use far less CPU time to produce useful results. |
|