Diversity clustering
Fingerprints, BitBIRCH clustering and representative picking — cut the library before you spend a docking run on it.
AMDockVS is a distributed platform for structure-based drug discovery. Protein–ligand virtual screening today, with a docking layer built to take other campaign types as they land. Two flows share one project model: VS for libraries in the hundreds of thousands on a single workstation, HTPVS for millions, sharded across a local network or an HPC queue.
Beta · Python 3.12+ · MIT · built on MolSuite
Millions in, thousands out: every stage runs as shards across every worker you have, and only hits come back.
10⁵+
Ligands per VS project, one workstation
10⁶+
Ligands per HTPVS campaign, distributed
0
Brokers, daemons or servers to keep alive
Platform
Both share the same docking layer, the same engines and the same results table — what changes is whether the library lives in the project or streams past it.
VS
Ligands are rows in the project database: browsable, filterable, one click from a pose. Standardise, compute descriptors, cluster for diversity, dock a scope and read the results in the same table you imported into. The right shape for a few hundred thousand molecules on the machine under your desk.
HTPVS
The import splits the library into uniform shards and materialises nothing. One shard is one unit of work, so a campaign fans out over cores, a lab network or an HPC allocation, and only promoted hits ever become rows — which is what keeps a ten-million-molecule run from filling the database it was meant to stay out of.
The mode is read off the data, never declared: a project holds shards or it holds rows, and the interface follows. Docking engines, pocket detection, preparation and analysis are one layer below both — which is where other campaign types will plug in.
Tools
Fingerprints, BitBIRCH clustering and representative picking — cut the library before you spend a docking run on it.
Descriptors, activities, datasets and models. Train on what you have, score what you don’t, keep both in the project.
SMILES to 3D conformers, MMFF minimization, protonation states, tautomers and torsion trees, applied across the whole scope.
From sequence to structure: homology or AlphaFold models, protonation at your pH, waters removed and charges assigned.
P2Rank cavity prediction, residue selection, a co-crystal ligand, or a grid box you set by hand.
Vina 1.2.7, QuickVina2, GNINA — classical scoring and AI-based rescoring behind one interface.
Docking, re-docking and off-target docking against a second receptor — same inventory, same results table.
Scope a set, dock it, rank it, open the pose — all in the table you imported into.
Shards as work units, dispatched over cores, a lab network or an HPC queue. Crashes resume; only hits are promoted.
Contact maps and 2D interaction diagrams for the poses that survive — figure-ready, from the app or the CLI.
Documentation
The application: install, project modes, engines and grids, pocket detection, QSAR, and how a screening campaign is set up and resumed.
The stack underneath: in-process orchestration, shared Qt components, and contact maps. Explore the source repositories in the MolSuite organization.
Publications & citation
The platform paper is in preparation. Until it lands, cite the software release and the predecessor tool your protocol builds on.
Placeholder — send me the DOI, journal and release version and I’ll set the real entries.
BibTeX — placeholder
@software{amdockvs,
title = {AMDock-VS: a distributed platform
for virtual screening},
author = {Valdés-Tresanco, Mario S. and
Valdés-Tresanco, Mario E. and
Moreno, Ernesto and
Valiente, Pedro A.},
year = {——},
version = {——},
doi = {——},
url = {https://github.com/AMDockVS/AMDockVS}
}
Team
Mario S. Valdés-Tresanco, PhD
Author · maintainer
University of Medellin, Colombia
Mario E. Valdés-Tresanco, PhD
Author · maintainer
University of Calgary, Canada
Ernesto Moreno, PhD
Author
University of Medellin, Colombia
Pedro A. Valiente, PhD
Author
University of Toronto, Canada