Hacker News Reader

SETI wasn't a waste of time don't have any good alternatives to it. Folding@home on the other hand probably could have been a lot more useful if they were training transformers on their results instead of just doing pure protein dynamics simulations.

View on HN

Folding@Home predates the ML approaches to protein folding. Just because it has been eclipsed now doesn’t mean it wasn’t worthwhile.

Well, it predates the deep-learning ML approaches to protein folding like Alphafold. But people have been applying ML methods to protein folding for decades -- that has long been the two camps in the protein-folding bioinformatics world -- whether the best way to get structures was to simulate what the protein was actually doing or if there was a computational shortcut.

The goal of protein folding simulations like Folding@Home is not to predict 3D structures - it's to understand how folding actually works, and why it sometimes doesn't work. When FAH came out it was already very obvious that there were good computational shortcuts to predicting the end state (the Rosetta approach), but those don't tell you very much about the physical process. Different questions call for different approaches.

Reply on news.ycombinator.com