AutoDock-GPU

What is AutoDock-GPU?

AutoDock-GPU is a GPU-accelerated implementation of AutoDock4, one of the most widely-cited docking programs in computational drug discovery. It predicts how small molecules bind to protein targets using the classic AutoDock4 physics-based force field, but runs up to 350x faster than single-threaded AutoDock4 by leveraging NVIDIA GPU parallelization.

The key distinction from AutoDock Vina is the scoring function: AutoDock-GPU uses the original AutoDock4 force field with explicit terms for van der Waals interactions, hydrogen bonding, electrostatics, and desolvation. This physics-inspired approach behaves differently than Vina's empirically-optimized scoring, particularly for metal-containing binding sites and systems with complex electrostatics.

AutoDock-GPU is developed by the Forli Lab at Scripps Research and is open-source under the GPL license. ProteinIQ handles all the complexities of grid map generation and file preparation—just provide a protein structure and ligand SMILES to get started.

How does AutoDock-GPU work?

AutoDock-GPU combines pre-calculated grid maps with the Lamarckian Genetic Algorithm (LGA) to efficiently search conformational space on GPUs.

The AutoDock4 scoring function

The scoring function estimates binding free energy ($\Delta G$) through five physics-based terms:

$\Delta G = \Delta G_{vdW} + \Delta G_{hbond} + \Delta G_{elec} + \Delta G_{desolv} + \Delta G_{tor}$

• Van der Waals ($\Delta G_{vdW}$): Lennard-Jones 12-6 potential modeling steric interactions
• Hydrogen bonding ($\Delta G_{hbond}$): 12-10 potential with a directional term favoring optimal H-bond geometry
• Electrostatics ($$): Coulombic interactions with a distance-dependent dielectric

These interaction energies are pre-calculated by AutoGrid at 0.375 Å resolution and stored as 3D grids. During docking, energies are rapidly looked up via trilinear interpolation rather than computed pairwise—this is what enables ~250 million scoring function evaluations per docking job.

The Lamarckian Genetic Algorithm

The LGA is a hybrid global-local search strategy:

1. Population initialization: Random poses (position, orientation, torsion angles) are generated
2. Genetic evolution: New poses are created through crossover and mutation of parent poses
3. Local refinement: ~6% of the population undergoes local search optimization each generation
4. Lamarckian inheritance: Improved poses from local search re-enter the population (unlike Darwinian evolution where acquired traits aren't inherited)
5. Termination: Search stops when maximum evaluations or generations are reached

Multiple independent LGA runs (controlled by the "Number of LGA runs" parameter) explore different regions of conformational space. The best pose across all runs is returned.

Local search methods

AutoDock-GPU offers two local search optimizers:

ADADELTA (gradient-based): Uses scoring function gradients to descend toward energy minima. Converges faster for flexible ligands (8+ rotatable bonds) and typically requires 6-23x fewer evaluations than Solis-Wets. We recommend this for most applications.

Solis-Wets (random): Applies random perturbations to pose variables, accepting improvements probabilistically. Better for small, rigid ligands (≤7 rotatable bonds) where gradient information is less beneficial.

GPU parallelization

AutoDock-GPU exploits three levels of parallelism:

• High-level: Multiple LGA runs execute simultaneously across GPU threads
• Medium-level: Individuals within a genetic generation are processed in parallel
• Low-level: Fine-grained tasks (rotation, scoring) are distributed across work-items

This hierarchical approach achieves 30-350x speedup depending on the GPU and system size. A single GPU can screen 2,000-40,000 ligands per day.

Input requirements

Protein (Receptor): PDB file or RCSB PDB ID. ProteinIQ automatically adds hydrogens and converts to PDBQT format. For best results, first use PDB Fixer to repair missing residues.

Ligand: SMILES string, SDF, MOL, or PDBQT file. SMILES is the simplest option for small molecules. For ligands containing metals, use GNINA instead—AutoDock-GPU's parameterization doesn't handle metal coordination in ligands.

Docking parameters

Core settings

Number of LGA runs: Independent genetic algorithm runs per docking job. More runs explore more conformational space but increase runtime. Use 20 (default) for screening, 50-100 for thorough analysis of lead compounds.

Local search method: Choose ADADELTA for flexible ligands (8+ rotatable bonds) or Solis-Wets for small, rigid molecules. When unsure, use ADADELTA—it's faster for complex ligands and comparable for simple ones.

Advanced settings

Max evaluations per run: Scoring function evaluations before terminating each LGA run. Higher values allow more thorough search at the cost of runtime. The default (2,500,000) balances speed and accuracy.

Grid spacing: Resolution of pre-calculated energy maps in Angstroms. Smaller values (0.2-0.3) increase accuracy but require more memory. The default (0.375 Å) is the AutoDock standard.

Search space: By default, ProteinIQ searches the entire protein surface (blind docking). For known binding sites, enable manual mode and specify center coordinates and grid dimensions. A focused search box dramatically reduces runtime.

Understanding the results

Binding affinity interpretation

AutoDock-GPU reports binding affinity in kcal/mol. More negative values indicate stronger predicted binding:

AutoDock4 and Vina scoring functions are calibrated differently—don't directly compare affinities between the two programs. Use AutoDock-GPU scores for ranking compounds docked with AutoDock-GPU only.

Output files

Results include multiple binding poses ranked by affinity. Each pose is provided in PDBQT format, which can be visualized in the 3D viewer or downloaded for analysis in molecular visualization software.

When to use AutoDock-GPU vs alternatives

AutoDock-GPU is the best choice when you specifically need AutoDock4 scoring (e.g., for consistency with published results) or when screening large compound libraries where GPU acceleration provides the biggest advantage.

Comparison to other docking tools

Frequently asked questions

What's the difference between AutoDock-GPU and AutoDock Vina?

AutoDock-GPU uses the original AutoDock4 physics-based scoring function with a Lamarckian Genetic Algorithm search. AutoDock Vina uses an empirically-optimized scoring function with BFGS local search. Vina is simpler to use (no grid maps needed), while AutoDock-GPU provides the classic AD4 scoring that some workflows require. Both are from the same research group at Scripps.

Why does AutoDock-GPU need grid maps?

Grid maps pre-calculate interaction energies between each ligand atom type and the receptor at every point in space. This upfront cost (handled automatically by ProteinIQ) enables extremely fast scoring during the docking search—AutoDock-GPU can perform ~250 million scoring evaluations per job by looking up cached values rather than computing pairwise interactions.

Is AutoDock-GPU faster than AutoDock Vina?

For individual dockings, they're comparable (1-5 minutes). AutoDock-GPU's advantage emerges in high-throughput screening: once grid maps are generated for a receptor, subsequent ligands dock very quickly. GPU parallelization enables screening 2,000-40,000 compounds per day per GPU.

Which local search method should I use?

Use ADADELTA (default) for most cases. It uses gradient information to converge faster, especially for flexible ligands with 8+ rotatable bonds. Switch to Solis-Wets only for very rigid molecules (≤7 rotatable bonds) or if you need to reproduce classic AutoDock4 behavior.

Can I dock peptides or macrocycles?

AutoDock-GPU handles molecules up to about 32 rotatable bonds effectively. For larger peptides or macrocyclic compounds, the search space becomes extremely large. Consider reducing the number of flexible bonds, using DiffDock which handles flexibility differently, or preparing constrained PDBQT files externally.

How do I interpret very negative binding affinities?

Scores below -12 kcal/mol may be scoring artifacts rather than true predictions. Common causes include ligands with many rotatable bonds or binding poses with unphysical geometry. Visually inspect such poses in the 3D viewer and validate with alternative methods.

When should I define manual search space coordinates?

Define manual coordinates when you know the binding site location. This focuses the search, reduces runtime, and often improves results. Use the protein's center of mass or coordinates from a co-crystallized ligand as a starting point. For blind docking (unknown binding site), use the default auto mode.

Related tools

• AutoDock Vina - Faster docking with empirical scoring
• GNINA - CNN-enhanced docking for improved pose accuracy
• DiffDock - Diffusion-based blind docking
• Smina - Vina fork with custom scoring options
• PDB Fixer - Prepare proteins before docking
• ADMET-AI - Predict pharmacokinetic properties of docked compounds
• Lipinski's Rule of Five - Assess drug-likeness before virtual screening

References

Santos-Martins, D., Solis-Vasquez, L., Tillack, A.F., Sanner, M.F., Koch, A., Forli, S. (2021). Accelerating AutoDock4 with GPUs and Gradient-Based Local Search. Journal of Chemical Theory and Computation, 17(2), 1060-1073. DOI: 10.1021/acs.jctc.0c01006

Morris, G.M., Huey, R., Lindstrom, W., Sanner, M.F., Belew, R.K., Goodsell, D.S., Olson, A.J. (2009). AutoDock4 and AutoDockTools4: Automated docking with selective receptor flexibility. Journal of Computational Chemistry, 30(16), 2785-2791. DOI: 10.1002/jcc.21256