PoseBusters checks

PoseBusters tests whether a docked pose is physically plausible. It checks bond lengths and angles, flat rings, stereochemistry, clashes, the ligand's energy and its contacts with the protein. PoseBusters is installed with PocketVina. Docking does not run it unless you ask for it.

One pair

bust() checks the poses you picked with select():

pair = pocketvina.select(protein="1fbl", ligand="molecule1")              # the best pose
pair.bust()
pair = pocketvina.select(protein="1fbl", ligand="molecule1", mode="all")  # every mode of the best pocket
checks = pair.bust(workers=4)                                             # 4 processes at a time
checks[checks.pb_valid]                                                   # the modes that pass every check

bust() returns a table with one row per selected pose. Nothing is written to disk. The main columns:

• pb_valid: true if every check passed

• pb_status: valid, invalid, or why the pose was not checked

• failed_checks: the names of the checks that failed

• then one column per check: passed, failed, or empty if it does not apply

A whole run

pocketvina bust checks the poses of a whole results folder:

pocketvina bust --results results                                        # the best pose of every ligand
pocketvina bust --results results --select top_k --top-k 3 --overwrite   # the 3 best modes of every ligand
pocketvina bust --results results --select per_pocket --top-k 1 --protein 1fbl --overwrite

The checks are saved in the results folder, in tables/posebusters/. Proteins that were checked already are skipped, so you can run bust again after docking more proteins.

Option	Meaning	Default
--select best	the best pose of each ligand	best
--select top_k --top-k K	the K best modes of each ligand, over all pockets
--select per_pocket --top-k K	the K best modes of each ligand in each pocket
--config dock	PoseBusters configuration: dock, dock_fast, ...	dock
--workers N	worker processes	the CPUs, at most 16
--timeouts 30,300	seconds per pose for each try. A pose that takes longer is tried once more, then saved as timeout	30,300
--overwrite	check proteins again
--protein P	only this protein. Can be given more than once	all

You can also check while docking. Add --posebusters, or set posebusters=True in PocketVinaConfig. The CPUs then check each protein while the GPU docks the next one:

pocketvina dock --proteins example/proteins --ligands example/ligands.txt --out results --posebusters

In Python, check a whole run and read the results back:

from pocketvina import bust
from pocketvina.io.results import Results

bust.bust_results("results", select="best", workers=4)        # returns the protein ids checked
pb = Results("results").posebusters("1fbl")
pb[["ligand_id", "pocket", "mode", "pb_status", "failed_checks"]]

Outside Linux, put this code under if __name__ == "__main__": in a script.

Saved results

pb_status	Meaning
valid	every check passed
invalid	at least one check failed. failed_checks names them
timeout	still running after the last timeout
error	PoseBusters failed. The error column has the message
convert_failed	the pose could not be turned into a molecule
no_pose	the ligand has no pose in this protein

The columns pb_mask and pb_na_mask store each check as one bit: passed, or no result. Results("results").posebusters_checks("1fbl") returns the check names in bit order.

You can query the checks with DuckDB:

pocketvina query --results results "SELECT pb_status, count(*) AS n FROM posebusters GROUP BY pb_status"
pocketvina query --results results "SELECT protein_id, ligand_id, pocket, mode, failed_checks FROM posebusters WHERE NOT pb_valid LIMIT 20"