Foldit Lab Report #11: Deep Learning
Details of a new research paper that Foldit players contributed to! In it, scientists at the University of Washington and at Harvard developed a new, machine-learning based method for protein design. Watch our latest lab report for more.
NEWS
- How did Foldit help with the new deep learning paper?
PUZZLE UPDATES
- Coronavirus binders & anti-inflammatory design
- More symmetric design, and the problem of aggregation
DESIGN OF THE MONTH
- NinjaGreg made a sweet symmetric trimer!
See for yourself in the Design of the Month sandbox puzzle!
Good catch! This kind of geometry is problematic for all hydroxyl (-OH) groups, including the -OH at the tip of TYR.
interesting, this kinda answered a nagging question I had- in that I've wondered if someone is leveraging our "bad" designs as much as our "good" designs- and any designs we make we make just before we click 'reset'.
Sometimes I just like making patterns to play around. I would think these would be just as significant as a high scoring design because there was at least some intentional logic that went into it.
It's the same idea as how dictionary attacks are used in password cracking. If you are looking to crack a password, it helps to have a library of all words that exist, all passwords that have been previously used by anyone from any system, and common character substitutions (eg. l33t speak.) By using these patterns you are maximizing the value of everyone's work.
or any manipulations I make to a design with my mouse. It seems like these intentional movements would all be valuable data to use.
Are tyrosine triangles also problematic?