Using AutoDock GPU for large docking project

[KirkDCO]

I'm working on a project in which I would like to screen a very large number of structures with AutoDock. I have a good target structure with a well defined binding site to use for docking. The challenge is that I want to dock millions of candidate structures.

For such a task, what settings and optimizations should I consider for AutoDock-gpu? Obviously speed of critical importance, but I don't want to sacrifice quality of results.

[diogomart]

Just use the defaults, we optimized them for this goal. I would add -C 1 to calculate distance based interactions that are useful for filtering with ringtail https://ringtail.readthedocs.io/

[diogomart]

Be sure to pass a large number of ligands at once, that has a significant impact on runtime.

[KirkDCO]

Is the only way to pass more than one ligand at a time to include those on the command line? Can I supply a .pdbqt with multiple ligands in it?

…

On Wed, Jan 1, 2025 at 4:50 PM Diogo ***@***.***> wrote: Be sure to pass a large number of ligands at once, that has a significant impact on runtime. — Reply to this email directly, view it on GitHub <#288 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AFNZAYKFBQV5QP7KBQOQXHT2IR5KTAVCNFSM6AAAAABUO6NJ3OVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKNRXGE4DMNZTG4> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[atillack]

You can use the batch file feature -B (also described close to the end of the README.md file) for this purpose.

When you compiled with OVERLAP=ON (default since v1.6) ligand reading and results processing overlap with docking to maximize throughput. Also, when you have multiple GPUs you can run with -D all on all available GPUs simultaneously.

Additionally, I also added the ability to -B to be passed a list of ligand files, i.e. -B ligandsA/*.pdbqt. Just keep in mind that command line expansions are handled by the shell which limits the buffer space so it will be a good idea (also for other reasons like file system efficiency) to not have more than a couple tens of thousands of ligands in one folder. The output files in this mode will have the same base name as the original ligands but with dlg and/or xml endings.

[KirkDCO]

Ah, I didn't read the README.md thoroughly. 8^( Thank you for pointing out the batch file option.

Do I understand correctly, that if I have only one target and many ligands, the batch file can look like this?

./receptor1.maps.fld
./ligand1.pdbqt
Ligand 1
./ligand2.pdbqt
Ligand 2
./ligand3.pdbqt
Ligand 3

Thank you also for pointing out the command line expansion peril. With >500,000 structures, I will opt for the batch file.

[atillack]

Yes, also if you don't specify an output base name ("Ligand 1", etc.) it'll use the ligand pdbqt ones, so this is the "minimum viable batch file":

receptor.maps.fld
ligand1.pdbqt
ligand2.pdbqt
ligand3.pdbqt

(which also is equivalent to -M receptor.maps.fld -B ligand1.pdbqt ligand2.pdbqt ligand3.pdbqt)

[KirkDCO]

Excellent! I'll give this a try.

I plan to gather these valuable tips and other details I find into a cookbook style document. I'll share it once I have something of substance, if there's interest in adding it to the documentation.

[diogomart]

Yes, that would be awesome!

[KirkDCO]

Just a quick follow up for anyone that finds this thread.

I just a timing test docking 1000 structures. Running AutoDock-GPU repeatedly with one structure at a time and on one GPU took about 44 minutes.

Using the -B batch file option described above, reduced that by about 50%, and adding -D all (with 2 RTX 4090 GPUs available) reduced it by another 50%. Now 1000 structures takes about 9 minutes. !!!

Thanks for these great tips @diogomart and @atillack !!

[diogomart]

Thanks for sharing the speedup you observed!