SMINABeta

What is SMINA?

SMINA is a fork of AutoDock Vina developed by David Koes at the University of Pittsburgh. It extends Vina with comprehensive support for custom scoring functions, dramatically faster minimization, and enhanced workflows for scoring function research.

Published in 2013, SMINA was designed to address limitations in Vina's scoring function development capabilities. The tool enables researchers to define custom scoring functions with user-specified weights, making it valuable for both standard docking and advanced scoring research.

SMINA also includes the Vinardo scoring function, an optimized variant that consistently outperforms Vina's default scoring in pose prediction, ranking, and virtual screening benchmarks.

How does SMINA work?

SMINA inherits Vina's core search algorithm while extending its scoring capabilities. The search uses Iterated Local Search with BFGS quasi-Newton optimization, exploring the conformational space through random mutations of position, orientation, and torsion angles.

Scoring functions

SMINA provides three scoring options:

Default (Vina) uses the original AutoDock Vina scoring function with Gaussian steric terms, hydrogen bonding, hydrophobic interactions, and a rotatable bond penalty. This is well-validated and suitable for most docking applications.

Vinardo removes Vina's problematic second Gaussian term that contributed 58% of binding energy despite creating artifacts. Vinardo uses optimized atomic radii and simplified terms, achieving 89.7% top-1 docking success versus Vina's 80.5% on the CASF-2013 benchmark.

Custom scoring allows researchers to define their own scoring functions using 58 available interaction terms, including electrostatics, desolvation, Lennard-Jones 4-8 potentials, and simple property counts.

Available scoring terms

SMINA's custom scoring supports these term types:

• Steric: Gaussian functions (gauss), repulsion potentials
• Chemical: Hydrogen bonding, hydrophobic contacts, non-hydrophobic contacts
• Electrostatic: Distance-dependent electrostatic terms
• Desolvation: AutoDock 4-style solvation model
• Van der Waals: Lennard-Jones 4-8 formulation
• Property counts: Torsion counts, hydrophobic atom counts

Each term accepts parameters for offset, width, cutoff, and exponent, enabling fine-grained control over the scoring function behavior.

Docking modes

SMINA provides four operational modes:

The Minimize only mode is particularly valuable for pose refinement workflows, achieving 10-20x speedup over full docking by skipping global search.

Input requirements

Protein (Receptor)

Upload a PDB file or enter an RCSB PDB ID. The protein should be prepared with hydrogens added and missing residues fixed. Use PDB Fixer for automated preparation.

Ligand

Enter a SMILES string for small molecules, or upload an SDF, MOL, or PDBQT file. SMINA automatically generates 3D coordinates and calculates partial charges via OpenBabel.

For complex ligands (macrocycles, peptides, >150 atoms), pre-prepared PDBQT files are recommended to ensure proper handling.

Docking parameters

Scoring function

• Default (Vina): Standard Vina scoring, well-validated for general docking
• Vinardo: Optimized for pose prediction accuracy, recommended for structure-based drug design
• Custom scoring file: Define your own scoring weights for research applications

Docking mode

• Full docking: Complete global search with local optimization
• Minimize only: Fast pose refinement (10-20x faster than full docking)
• Score only: Evaluate a pose without any optimization
• Local only: Optimize from the input starting pose

Search parameters

• Exhaustiveness: Number of independent docking runs. Higher values (16-32) improve search coverage for difficult targets
• Number of poses: Maximum binding poses to return. SMINA supports up to 50 poses, versus Vina's limit of 20
• Energy range: Maximum energy difference from the best pose for inclusion. Only poses within this range are returned
• RMSD filter: Minimum RMSD between poses to reduce redundancy

Search space

The search box defines where SMINA explores ligand binding:

• Auto: Automatically calculates a box encompassing the entire protein surface
• Autobox: Derives the search box from a reference ligand (useful when binding site is known from a co-crystal structure)
• Manual: Specify exact center coordinates and box dimensions

We recommend using autobox when a reference ligand structure is available, as this focuses the search and improves both speed and accuracy.

Output options

• Per-atom interaction energies: Outputs energy contributions for each atom, useful for analyzing which parts of the ligand contribute most to binding

Understanding the results

Binding affinity

Affinity is reported in kcal/mol. More negative values indicate stronger predicted binding:

Vinardo scores may differ slightly from Vina due to the modified scoring function. Both use the same scale and interpretation guidelines.

Pose ranking

SMINA ranks poses by predicted binding affinity. The top-ranked pose represents the most favorable binding mode, but examining multiple poses is recommended. Alternative binding modes within 1-2 kcal/mol of the best score may be biologically relevant.

Validation metrics

On the CASF-2013 benchmark, Vinardo achieves:

• Docking power: 89.7% success rate (pose RMSD < 2Å from crystal structure)
• Screening power: Identifies true binders in top 1% for 80% of proteins
• Scoring correlation: r = 0.601 between predicted and experimental binding affinities

Comparison to other docking tools

SMINA is ideal when you need custom scoring functions, fast minimization, or more than 20 poses. For standard docking with known scoring, Vina or GNINA work equally well.

Custom scoring function syntax

Custom scoring functions are defined as weighted sums of interaction terms. Each line specifies a weight and term with parameters:

-0.035579 gauss(o=0,_w=0.5,_c=8)
-0.005156 gauss(o=3,_w=2,_c=8)
0.840245 repulsion(o=0,_c=8)
-0.035069 hydrophobic(g=0.5,_b=1.5,_c=8)
-0.587439 non_dir_h_bond(g=-0.7,_b=0,_c=8)
1.923 num_tors_div

Parameters include:

• o: Offset from optimal distance
• w: Gaussian width
• c: Distance cutoff
• g: Good distance threshold
• b: Bad distance threshold

Detailed term documentation is available in the SMINA GitHub repository.

Best practices

• Use Vinardo for pose prediction: Vinardo consistently outperforms default Vina scoring in docking benchmarks
• Define search space when possible: Using autobox with a reference ligand improves both speed and accuracy
• Use minimize mode for refinement: When you have a reasonable starting pose, minimize mode is 10-20x faster than full docking
• Increase exhaustiveness for difficult targets: Flexible ligands or large binding sites benefit from exhaustiveness values of 16-32