AutoDock Vina Batch Docking: How to Screen a Ligand Library Without Losing Traceability

The official AutoDock Vina documentation defines batch mode as docking ligands one after another and exposes the --batch and --dir options for that purpose [1]. This is different from simultaneous multiple-ligand docking, where more than one ligand participates in the same docking calculation [2]. Both capabilities exist, but they answer different scientific questions.

This guide concerns the common virtual-screening situation: one prepared receptor, one frozen search protocol, and a library of independent small molecules that must remain identifiable from import through pose review.

A directory containing prepared ligands and Vina outputs may be enough to start a calculation, but it does not automatically preserve the scientific record. Filenames can collide, one SDF may contain several records, failed molecules can disappear from the output directory, and a rerun may silently use a different box or seed.

The screening denominator therefore begins with every submitted source record, not only the ligands that produced a score. A defensible summary distinguishes at least: received, rejected at import, held during preparation review, eligible for docking, attempted, completed, failed, cancelled, and retained for pose review.

Batch execution multiplies a protocol; it does not validate it. Establish the receptor state, binding-site rationale, preparation policy, scoring function, search box, and decision rule before the production library is exposed to them. Use target-relevant controls where available and keep pose recovery separate from screening discrimination.

The detailed molecular docking workflow validation guide covers redocking, controls, enrichment, robustness, and a defensible go/no-go decision. The AutoDock Vina grid-box guide explains how to document center, size, and exhaustiveness before they become batch-wide settings.

Do not wait for PDBQT filenames to become the identifiers. Give each submitted molecular record a stable internal ID while its original source filename, record index, external identifier, and checksum are still available. If one SDF contains several molecules, each record needs its own row. Meeko explicitly supports SD files with multiple molecules and can write separate PDBQT files for them [5].

Keep the scientific identity separate from the display name. Two ligands may share a title, and the same compound may be intentionally represented by several protonation, tautomeric, or stereochemical states. The batch manifest must be able to express both situations without overwriting a result.

Preparation failures are part of the screening result. Parse the intended record, resolve its chemical graph and docking state, generate a usable 3D structure when needed, and verify the PDBQT before marking it eligible. Meeko's current ligand-preparation documentation requires explicit hydrogens and 3D coordinates for its ordinary input path [5].

The SDF preflight checklist separates record integrity, chemical graph, docking state, and prepared-artifact checks. The broader protein and ligand preparation guide explains how those decisions fit around the receptor and ligand workflow.

Use the same prepared receptor, box, scoring function, search settings, and reporting rule for every ligand that belongs to one comparison. The official Vina tutorial shows a configuration file for the box and explains that higher exhaustiveness increases the computational effort used by the search [3]. The Vina FAQ further explains that exhaustiveness controls the number of independent search runs and that larger search spaces can require more effort and longer execution [4].

Record the random seed. Vina's search is stochastic, and its documentation states that the same seed, inputs, and parameters are required for exact reruns [4]. A fixed seed supports replay; it does not by itself demonstrate robustness. If the validation plan requires repeated seeds, define those repeats before looking at the production ranking and treat each repeat as a linked run.

Run a small, prespecified pilot that spans the chemistry and flexibility expected in the library. The purpose is operational: verify input parsing, preparation evidence, box containment, runtime behavior, output naming, failure capture, pose parsing, and export. It is not permission to select parameters that make the pilot compounds look favorable.

If the pilot reveals a protocol defect, correct and revalidate the protocol, assign a new version, and restart the comparison. Do not merge results from incompatible protocol versions into one ranking without a justified harmonization method.

AutoDock Vina's command-line batch mode accepts repeated --batch arguments and writes outputs beneath the directory supplied with --dir [1]. The documentation adds an index when duplicate ligand filenames would collide [1]. That prevents one overwrite, but it does not replace the manifest: an indexed filename still needs to map back to a stable ligand ID and source record.

Report counts at each gate. A statement such as "100 submitted, 87 eligible, 84 completed, and 3 docking failures" is only an illustrative reporting pattern, not a result from this article. The important point is that the three failures and thirteen pre-docking exclusions remain visible rather than disappearing from an 84-row score table.

Vina reports ranked modes, affinity values, and RMSD lower and upper bounds relative to the best generated mode; these values do not establish experimental binding or pose correctness. Review whether a pose occupies the intended pocket, respects the protocol boundary, avoids obvious structural conflicts, and preserves the interactions relevant to the hypothesis.

The guide to AutoDock Vina affinity, RMSD, and pose selection explains those output columns and the distinction between generated-pose RMSD bounds and RMSD to an experimental reference. Keep the selected-pose rationale attached to the ligand and batch record.

MolNexus 0.1.1 is BioChemIntelli's local Windows desktop workflow for receptor and ligand preparation, interaction-box setup, AutoDock Vina execution, pose review, export, and persistent job history. The current build accepts one or many supported ligand sources, prepares and reviews them, and can place up to 100 eligible ligands in one job. It then processes those ligands one at a time against one receptor while retaining per-ligand success or failure, poses, logs, preparation evidence, the box, Vina parameters, CPU allocation, and engine identity.

That is product behavior verified in the current application and tests. It is not evidence that MolNexus makes Vina more accurate, that every submitted chemistry is supported, or that a 100-ligand job has a guaranteed duration. The current development build integrates AutoDock Vina 1.2.7, whose official release record includes batch-directory support [6].

The documented case began with eight prespecified ligand files from the original YuelDesign dataset [7] and an explicitly prepared apo-CDK2 receptor derived from PDB 4EK3 [8]. The current preflight produced two Ready ligands, three Review outcomes, and three source-file rejections. Only the two Ready ligands entered this job. Preserving the other six outcomes is exactly what keeps the apparent two-of-two completion from being mistaken for the original submission denominator.

A reliable batch is not defined by how many Vina processes finished. It is defined by whether every submitted molecule, preparation decision, protocol parameter, terminal state, pose, and export remains connected. Validate first, freeze the protocol, preserve the full denominator, and inspect poses before turning a score table into a shortlist.

If the immediate problem is keeping those relationships visible, the next useful comparison is the guide to organizing receptors, ligands, poses, and job history. If the protocol itself is not yet defensible, return to the workflow validation guide before scaling.