AutoDock-GPU
AutoDock-GPU copied to clipboard
ccsb-scripps
AutoDock GPU Batch Docking Result Unreproducible
Dear Developers,
As always, thanks for the amazing program. I am running the latest AutoDock GPU on a Ubuntu system with 3 Nvidia A40. I complied the AutoDock GPU with overlap = 'on' and NUMWI=256.
Because I am trying to do a multi-stage docking, I first docked 120000 ligands with the following command: autodock_gpu_256wi -B firstBatch.txt -x 0 --nev 3000000 -E 20000000 -n 50 -D all. And I selected the top 1% of the ligands from the first batch based on the docking scores and ran the following command: autodock_gpu_256wi -B secondBatch.txt -x 0 --nev 3000000 -E 20000000 -n 100 -D all, where I changed the # of runs.
The top ligands in the first batch were the following:
Ena116776: -14.49
Ena525549: -14.23
Ena613695: -14.10
Ena359547: -14.06
Ena577608: -14.02
Ena702299: -13.91
Ena810220: -13.84
Ena195173: -13.83
Ena950965: -13.81
Ena412239: -13.75
For the second batch, the top ligands were the following:
Ena387979: -11.46
Ena615600: -11.27
Ena998052: -11.25
Ena102995: -11.20
Ena290373: -11.05
Ena285926: -11.02
Ena1213054: -11.00
Ena163077: -10.99
Ena484661: -10.98
Ena1213021: -10.97
It seemed very strange to me that, after increasing the # of runs from 50 to 100, the lowest energy among second batch docking simulations was only -11.46, which showed a great difference from the first batch. I understand that of course variation occurs due to randomness in searching and increase of runs, but how come a subset from the first batch has no ligand with energy lower than -12 after the # of runs was increased to 100, whereas the first batch had many? So I compared the performance of Ena116776 in the first batch and the second batch:
first batch run result for Ena116776:
_______________________________________________________________________
| | | | | |
Rank | Sub- | Run | Binding | Cluster | Reference | Grep
| Rank | | Energy | RMSD | RMSD | Pattern
_____|______|______|___________|_________|_________________|___________
1 1 12 **-14.49** 0.00 101.46 RANKING
1 2 19 -14.46 0.19 101.54 RANKING
1 3 36 -12.56 0.95 101.93 RANKING
1 4 34 -12.40 0.80 101.76 RANKING
1 5 28 -11.88 1.64 101.37 RANKING
1 6 9 -11.57 1.40 101.21 RANKING
2 1 39 -12.36 0.00 101.61 RANKING
2 2 22 -12.30 0.38 101.59 RANKING
2 3 3 -12.04 0.71 101.61 RANKING
2 4 18 -11.92 0.88 101.93 RANKING
2 5 6 -11.91 1.92 101.65 RANKING
2 6 17 -11.90 0.91 101.98 RANKING
2 7 49 -11.80 0.89 101.97 RANKING
2 8 37 -11.76 0.83 101.75 RANKING
2 9 41 -11.70 1.30 101.62 RANKING
2 10 20 -11.38 0.80 101.70 RANKING
3 1 24 -12.10 0.00 100.36 RANKING
4 1 46 -11.95 0.00 101.94 RANKING
4 2 7 -11.41 1.48 101.39 RANKING
4 3 33 -10.75 1.74 101.79 RANKING
4 4 2 -10.67 1.71 101.23 RANKING
5 1 13 -11.85 0.00 103.21 RANKING
5 2 47 -11.27 1.48 103.81 RANKING
5 3 32 -11.21 1.31 103.66 RANKING
6 1 4 -11.69 0.00 101.47 RANKING
6 2 27 -11.18 1.47 102.11 RANKING
7 1 38 -11.65 0.00 102.91 RANKING
7 2 31 -11.55 0.31 102.84 RANKING
7 3 48 -11.05 0.81 102.96 RANKING
8 1 1 -11.60 0.00 103.68 RANKING
9 1 50 -11.53 0.00 102.75 RANKING
9 2 10 -10.89 1.63 102.78 RANKING
10 1 25 -11.47 0.00 102.47 RANKING
10 2 40 -11.44 1.32 101.67 RANKING
11 1 11 -11.35 0.00 105.73 RANKING
11 2 26 -11.01 1.88 105.63 RANKING
11 3 5 -10.71 1.73 106.89 RANKING
11 4 23 -10.43 1.85 106.99 RANKING
11 5 35 -9.62 1.79 106.53 RANKING
12 1 15 -10.77 0.00 104.87 RANKING
13 1 30 -10.65 0.00 104.12 RANKING
14 1 43 -10.60 0.00 104.61 RANKING
15 1 45 -10.59 0.00 106.43 RANKING
16 1 44 -10.30 0.00 107.35 RANKING
16 2 14 -10.10 1.93 107.62 RANKING
17 1 16 -10.17 0.00 105.66 RANKING
18 1 8 -9.84 0.00 107.41 RANKING
18 2 21 -9.49 1.01 107.53 RANKING
18 3 42 -9.43 0.62 107.38 RANKING
18 4 29 -9.35 1.30 107.64 RANKING
Run time 1.122 sec
Idle time 0.767 sec
second run result for Ena116776:
_______________________________________________________________________
| | | | | |
Rank | Sub- | Run | Binding | Cluster | Reference | Grep
| Rank | | Energy | RMSD | RMSD | Pattern
_____|______|______|___________|_________|_________________|___________
1 1 72 -8.89 0.00 100.60 RANKING
1 2 87 -7.80 1.83 101.92 RANKING
2 1 71 -8.82 0.00 101.58 RANKING
2 2 16 -8.76 0.25 101.64 RANKING
2 3 37 -8.08 1.96 101.73 RANKING
2 4 57 -8.07 1.96 101.76 RANKING
2 5 44 -8.04 1.05 102.05 RANKING
2 6 28 -7.79 1.78 101.86 RANKING
3 1 48 -8.37 0.00 102.97 RANKING
3 2 41 -8.13 0.40 102.89 RANKING
3 3 46 -7.98 0.76 102.98 RANKING
3 4 49 -7.29 1.42 102.41 RANKING
4 1 50 -8.21 0.00 103.20 RANKING
5 1 92 -8.03 0.00 103.63 RANKING
5 2 43 -7.90 0.82 103.92 RANKING
6 1 15 -7.84 0.00 104.44 RANKING
6 2 63 -7.73 0.54 104.27 RANKING
7 1 76 -7.75 0.00 106.96 RANKING
7 2 59 -7.66 0.17 106.95 RANKING
7 3 96 -7.65 0.23 106.92 RANKING
7 4 83 -7.58 1.06 107.40 RANKING
7 5 70 -7.55 1.08 107.44 RANKING
7 6 33 -7.50 1.10 107.44 RANKING
7 7 1 -7.41 1.38 106.51 RANKING
7 8 9 -7.33 1.28 106.98 RANKING
7 9 8 -7.32 1.26 106.92 RANKING
7 10 39 -6.83 1.16 107.03 RANKING
8 1 52 -7.66 0.00 101.41 RANKING
9 1 86 -7.59 0.00 105.48 RANKING
9 2 73 -7.40 1.59 105.18 RANKING
9 3 94 -7.32 1.57 104.51 RANKING
10 1 13 -7.55 0.00 105.86 RANKING
11 1 62 -7.47 0.00 106.12 RANKING
12 1 5 -7.46 0.00 106.21 RANKING
12 2 81 -7.45 0.89 106.52 RANKING
13 1 30 -7.43 0.00 105.77 RANKING
13 2 29 -6.91 1.65 105.86 RANKING
14 1 34 -7.40 0.00 103.92 RANKING
14 2 82 -7.37 1.50 104.72 RANKING
14 3 53 -7.36 1.52 104.75 RANKING
15 1 47 -7.23 0.00 107.41 RANKING
15 2 7 -7.20 0.32 107.44 RANKING
15 3 56 -6.98 0.74 107.59 RANKING
16 1 68 -7.16 0.00 103.93 RANKING
17 1 55 -7.16 0.00 108.21 RANKING
18 1 54 -7.15 0.00 100.80 RANKING
19 1 69 -6.87 0.00 107.54 RANKING
19 2 91 -6.77 0.13 107.50 RANKING
19 3 89 -6.76 1.31 107.11 RANKING
19 4 84 -6.76 0.13 107.55 RANKING
19 5 88 -6.73 0.25 107.45 RANKING
19 6 64 -6.60 0.44 107.69 RANKING
19 7 40 -6.56 1.18 106.93 RANKING
19 8 25 -6.55 0.54 107.77 RANKING
19 9 45 -6.38 1.53 107.76 RANKING
19 10 26 -6.37 0.31 107.62 RANKING
19 11 12 -6.31 1.87 107.77 RANKING
19 12 66 -6.30 1.87 107.75 RANKING
19 13 90 -6.28 1.88 107.72 RANKING
19 14 23 -6.24 1.89 107.83 RANKING
19 15 6 -6.16 1.23 107.34 RANKING
19 16 35 -6.09 1.18 107.38 RANKING
19 17 61 -6.03 1.21 106.92 RANKING
19 18 78 -5.85 1.24 107.53 RANKING
19 19 3 -5.82 1.37 107.84 RANKING
19 20 38 -5.81 1.67 107.94 RANKING
19 21 97 -5.73 1.46 107.57 RANKING
20 1 67 -6.64 0.00 107.87 RANKING
21 1 93 -6.38 0.00 108.10 RANKING
22 1 98 -6.31 0.00 109.11 RANKING
23 1 10 -6.22 0.00 107.43 RANKING
23 2 60 -6.18 1.09 106.95 RANKING
23 3 27 -6.13 0.70 107.60 RANKING
23 4 74 -5.84 1.68 106.48 RANKING
24 1 31 -6.20 0.00 108.53 RANKING
24 2 2 -6.14 1.38 107.89 RANKING
24 3 32 -6.14 0.17 108.50 RANKING
24 4 95 -6.11 1.37 107.86 RANKING
24 5 85 -6.10 1.35 107.93 RANKING
24 6 65 -6.07 1.13 108.14 RANKING
24 7 17 -5.99 1.13 108.11 RANKING
24 8 36 -5.97 1.12 108.10 RANKING
24 9 20 -5.97 1.12 108.08 RANKING
24 10 100 -5.95 1.11 108.09 RANKING
24 11 51 -5.93 1.14 108.08 RANKING
24 12 58 -5.89 1.18 108.01 RANKING
24 13 42 -5.88 1.15 108.01 RANKING
24 14 22 -5.87 1.16 108.00 RANKING
24 15 19 -5.85 1.21 108.03 RANKING
24 16 99 -5.69 1.41 107.91 RANKING
25 1 24 -6.17 0.00 107.19 RANKING
25 2 11 -6.16 1.22 107.27 RANKING
25 3 14 -5.87 0.86 107.37 RANKING
26 1 79 -5.98 0.00 107.07 RANKING
27 1 4 -5.93 0.00 107.54 RANKING
28 1 75 -5.81 0.00 108.85 RANKING
29 1 21 -5.81 0.00 107.19 RANKING
29 2 77 -5.48 1.69 107.44 RANKING
30 1 18 -5.43 0.00 108.00 RANKING
30 2 80 -5.43 1.22 108.22 RANKING
Run time 1.341 sec
Idle time 0.000 sec
Again, I understand that variation exists due to randomness in searching. However, it can be clearly seen that among the 100 runs, the program failed to find any docked pose that is better than the worst docked pose (-9.35) in the first batch run result. I tried to run docking on this ligand (Ena116776) only, and I tried so many time, none of them could have a value comparable to -14 (I even tried to replicate by using the random seed in the first batch dlg file ). This does not make much sense to me because my search space is not large.
Number of grid points (x, y, z): 41, 41, 41
Grid center (x, y, z): 99.492000, 17.864000, -0.196000A
Grid size (x, y, z): 20.000000, 20.000000, 20.000000A
Grid spacing: 0.500000A
I am wondering whether batch docking mode with 3 cards has a different calculation scheme compared to docking ligands one by one. Please let me know if my problem description is not clear or detailed enough.
Hello, Thanks for reporting, this is unexpected, I'll try to reproduce the behavior in my next screening. A comment on the grid maps: 0.5 A spacing is too coarse, I recommend using the default or lower (<=0.375).
I found out that it could be due to the reason that I commented out some lines in the maps file. I will investigate it myself first and update here if I figured out what went wrong.