New metrics are better, but can improve....

Case number:845813-2010748
Topic:Game: Tools
Opened by:BootsMcGraw
Opened on:Tuesday, November 10, 2020 - 00:56
Last modified:Tuesday, November 10, 2020 - 03:34

I found that is was possible to design a protein that met the criteria for all three metrics as originally required, but it was so fragile that even breathing on the protein caused a metrics meltdown.

The new metrics should prove less frustrating to most players, since SASA of 1500 and DDG of -40 don't seem realistic. (Are they? Someone please explain how these thresholds were derived.)

The problem is: Is a SASA of 750 half as good as one of 1500? Is a SASA of 500 a third as good? 150 one tenth as good? At the lower end of the spectrum, these values are probably meaningless, and there is likely some threshold value below which the SASA doesn't matter... it ain't sticking to the target, and above which the protein can stick to the target, the higher the SASA, the better the fit.

The same general principal applies to DDG and SC.

I propose that the scoring for the three metrics be modified, starting with zero additional points at some minimum, non-zero value, and increasing based on how much more than the minimum the player can reach, up to some limit.

e.g. 0 points for SASA under 1000, 5 points for each increase of 1 A up to 1500, no additional increase if the SASA exceeds 1500.

Again, the same general principal applies to DDG and SC.

(Tue, 11/10/2020 - 00:56  |  1 comment)

infjamc's picture
User offline. Last seen 8 hours 11 min ago. Offline
Joined: 02/20/2009
Groups: Contenders

Another possible option to consider is to phase in the bonus quadratically. For example, if a SASA of 1500 is required to earn the full bonus, perhaps 750 should be worth (750 / 1500) ^ 2 = 1/4, and 500 should be worth (500 / 1500) ^ 2 = 1/9.


Developed by: UW Center for Game Science, UW Institute for Protein Design, Northeastern University, Vanderbilt University Meiler Lab, UC Davis
Supported by: DARPA, NSF, NIH, HHMI, Amazon, Microsoft, Adobe, Boehringer Ingelheim, RosettaCommons