Testing Sidechain Accuracy
Case number: | 845813-991787 |
Topic: | Game: Tools |
Opened by: | auntdeen |
Status: | Open |
Type: | Bug |
Opened on: | Wednesday, February 8, 2012 - 18:39 |
Last modified: | Monday, February 20, 2012 - 23:11 |
Since I still can't seem to fold my way out of a paper bag, at thom001's suggestion, I ran the following tests on Sidechain Accuracy, using puzzle 512. Although I have a September client available, it would not load that puzzle.
All testing was done from a reset of 512. Testing was done with Sidechain Flipper v2.2 (global script), stopped after the 6th segment was processed. The tests would strongly suggest that:
1) the current Main client is either not accessing the correct number of rotamers, or is not processing them properly (that score is lowest, other than October client).
2) for use during sidechain improvement scripts, the BSSA & ESSA sliders are apparently worthless - the best setting is, for both, at 1 rather than the defaults.
Although these results are only from the beginning of the script, it is likely that there would be more differences if run to end.
Current Main:
1=6 (improve)
2=1
3=9
4=1
5=20
6=18 (improve)
score = 9867
Current new Beta (run with BSSA & ESSA defaults):
1=10 (improve)
2=1 (says immediate improvement but didn't)
3=9
4=1
5=20
6=18 (improve)
score = 9870
Current new Beta (run with BSSA & ESSA at 1.0):
1=10 (improve)
2=1
3=19
4=1
5=37
6=28 (improve)
score = 9875
Current new Beta (run with BSSA & ESSA at 0):
1=1
2=1
3=1
4=1
5=1
6=1
score = no improvement
Current new Beta (run with BSSA at 1.0 ESSA at 0):
1=10 (improve)
2=1
3=1
4=1
5=1
6=1
score = 9856
Current new Beta (run with BSSA at 0 & ESSA at 1.0):
1=1
2=1
3=19
4=1
5=37
6=28 (improve)
score = 9873
November Beta:
1=6 (improve)
2=1 (says immediate improvement but didn't)
5=9
4=1
5=20
6=18 (improve)
score = 9873
October Beta:
1=6 (improve)
2=1
3=9
4=1
5=20,
6=19 (improve)
score = 9858
Status: Open » Open |
To make it a bit more clear for anyone who has not been staring at these results all morning...
The sequential numbers are the segment numbers the script was analyzing, and working on for improvement - the number after that is the number of rotamers found to work with.
So "6=19" is segment 6 - 19 rotamers found.
Status: Open » Open |
We have setup a Contest to try to debug this problem.
Join the Contest here:
http://fold.it/portal/smartlogin?destination=addcontestant/b888c7da7603ace705c43ed88032b36c
Using this starting model, please log the results of using particular scripts with different clients.
Thank you for your patience and hopefully we can find this bug soon!
Cant enter this contest using "old" main clients:
6522e6aa85e1b44433d58aff6c06017f
07e7a001798eb97fd84dd779e2b9fb3a
(binary version)
You've lost me beta. I'd love to help but I don't understand how. Which starting model? Model of what? Which scripts? Any? OK. Which clients? How can I identify which client I'm using? How would you know? Is there a client ID# somewhere?
Running the latest Beta for the past couple of days....acid tweeker v1.3 seems to have been reporting more sidechain positions than I thought were possible:
Below is from some scriptlogs
Puzzle 508
Sgmnt: 8 positions: 95
Sgmnt: 36 positions: 83
Puzzle 510
Sgmnt: 12 positions: 64
I can not find a scriptlog, but the other day I stopped acid tweeker because it was on a segment that had over 100 positions.
I went back to Puzzle 510b and ran acid tweeker v1.3 set to run just the sidechain_manipulate() function with the "guts" commented out. This modified script only reports the number of sidechains. Then I moved the sliders for wiggle accuracy and shake accuracy. Setting the shake accuracy to 1 made the script report more sidechain positions.. I think the Shake Accuracy function is not working properly
Shake Acc = 0 Shake Acc =1
Sgmnt: 2 positions: 13 37
Sgmnt: 3 positions: 3 3
Sgmnt: 4 positions: 5 13
Sgmnt: 5 positions: 17 17
Sgmnt: 6 positions: 5 13
Sgmnt: 7 positions: 15 43
Sgmnt: 8 positions: 10 28
Sgmnt: 10 positions: 7 19
Sgmnt: 11 positions: 8 8
Sgmnt: 12 positions: 29 75
Sgmnt: 14 positions: 17 49
Sgmnt: 15 positions: 7 19
Sgmnt: 16 positions: 5 13
Sgmnt: 17 positions: 5 5
Sgmnt: 18 positions: 10 28
Sgmnt: 19 positions: 25 25
Sgmnt: 20 positions: 4 4
Sgmnt: 22 positions: 10 10
Shake accuracy may not be the best name for the option, but the behavior is intended.
The real label would be "Add extra rotamers up to the Xth C-C bond* of every side chain". For example, alanine has only one C-C bond (the first one coming out from the backbone) whereas leucine has 2 C-C bonds (the last C-CH3 bond doesn't count).
So if you change ShAcc to 1, then alanine and leucine will both get extra rotamers that have more unusual positions around the first C-C bond.
If you change ShAcc to 2, then alanine and leucine will both get extra rotamers around the first C-C bond (because ShAcc 2 implies ShAcc 1), and leucine will get extra rotamers around the second C-C bond, but not alanine, because it doesn't have a second C-C bond.
Extra rotamers are generally beneficial. In the Baker lab, the default setting we use is ShAcc 2.
* The real term for this is Chi angle, and it refers to the dihedral angle around each C-C bond in the amino acid. Chi angles are typically numbered from 1-4, where 1 is the closest to the backbone.
Opening the General Options will show the Version of the client
I tried to run Acid Tweak on a puzzle overnight last night using the latest Beta release.
I expected it to be finished this morning but found it was still running.
The number of rotamers appears to be still increased from those normally seen prior to the introduction of any of the new sliders.
Previously the most you would see would be in the 40s, as I remember
Now I am getting in the hundreds for some sidechains
No details yet I'm afraid - will try to sort some
Thank you for doing this!
Fixing this problem is our top priority right now.