AMDockVS

AMDockVS exposes a composite runtime on top of MolSuite. The official entry point is:

from amdockvs import AMDockVSRuntime

The active public API is organised by domain:

• runtime.loader: ligand and receptor loading
• runtime.molecules: scopes, filters, sets and access to the general molecular inventory
• runtime.complexes: explicit receptor-ligand relations for redocking and rescoring
• runtime.chemistry: optional chemical preparation tools for ligands and receptors
• runtime.qsar: descriptors, activities, datasets and QSAR models
• runtime.pockets: P2Rank installation and cavity prediction
• runtime.docking: Meeko preparation, grids, docking and result queries

Installation

Python >= 3.12. The MolSuite packages and PyMOL (the pymol-open-source alpha wheels, via ms_components) are installed automatically from PyPI:

python3.12 -m venv .venv
source .venv/bin/activate
pip install AMDockVS

Start the GUI:

amdockvs

Optional extras

pip install "AMDockVS[all]"   # or pick one: bblean, pockets, receptors

Non-pip dependencies (optional)

P2Rank 2.5.1 is installed on demand from Molecule Tools → Pocket Detection. AMDockVS first reuses ~/Downloads/p2rank_2.5.1.tar.gz if it exists, and otherwise downloads the official distribution and verifies its checksum. Java 17 is likewise installed on demand into the active Conda environment, after a dry-run check that the transaction will not replace Python, RDKit, PyMOL, PySide, NumPy, Vina or Meeko. P2Rank is not redistributed inside the application.

The same operation is available from the backend:

runtime.pockets.install_p2rank()
job_id = runtime.pockets.predict(
    receptor_ids=[receptor_id],
    profile="default",
    threads=1,
)
runtime.wait_for_job(job_id)
sites = runtime.pockets.list_predictions(receptor_id=receptor_id)

Minimal workflow

from pathlib import Path

from amdockvs import AMDockVSRuntime

runtime = AMDockVSRuntime()
try:
    runtime.create_project(name="demo", folder=Path("./demo_project"))

    load_jobs = [
        *runtime.loader.load_ligands(["./ligands.sdf"], executor_name="thread"),
        *runtime.loader.load_receptors(["./receptor.pdb"], executor_name="thread"),
    ]
    runtime.wait_for_jobs(load_jobs)

    chemistry_job = runtime.chemistry.standardize_ligands(executor_name="thread")
    runtime.wait_for_job(chemistry_job)

    ligand_scope = runtime.molecules.select(role="ligand")
    receptor_scope = runtime.molecules.select(role="receptor", limit=1)

    descriptor_job = runtime.qsar.compute_descriptors(molecule_set=ligand_scope, executor_name="thread")
    runtime.wait_for_job(descriptor_job)

    receptor = next(runtime.molecules.stream(receptor_scope))
    runtime.docking.set_grid(
        receptor_id=int(receptor.id or 0),
        center=(12.0, 13.0, 10.0),
        size=(20.0, 20.0, 20.0),
    )

    prep_jobs = [
        runtime.docking.prepare_ligands(executor_name="thread"),
        runtime.docking.prepare_receptors(executor_name="thread"),
    ]
    runtime.wait_for_jobs(prep_jobs)

    receptor_set = runtime.molecules.create_set(receptor_scope, name="default_receptor_set")
    docking_job = runtime.docking.run(
        ligand_set=ligand_scope,
        receptor_set=receptor_set,
        executor_name="thread",
    )
    runtime.wait_for_job(docking_job)

    results = runtime.docking.list_results()
finally:
    runtime.shutdown()

Async semantics

• loader.*, qsar.compute_descriptors(), docking.prepare_*() and docking.run() submit jobs and return a job_id or a list[job_id].
• chemistry.* also submits jobs by default; use wait=True or runtime.wait_for_job(...) to block in notebooks.
• Waiting is explicit, through runtime.wait_for_job(...) or runtime.wait_for_jobs(...).
• runtime.docking.screen(...) is async by default too. Use wait=True only when you want a blocking composite helper.

Design notes

• stored_path keeps pointing at the base imported artifact.
• chemistry.* records a current chemical path in metadata; later steps consume it when present.
• Vina preparation is stored on amdock_molecules, without a separate per-state traceability layer.
• Inventory reads go through molsuite.query.project_rows(...); they do not depend on the private DataAccess helper.

Examples

• examples/pipeline_demo.py
• examples/runtime_quickstart.py

License

MIT — see LICENSE.