liveBook · Manning

Docking score

As explained in Build AI Drug Discovery Pipelines

A docking score is a numerical estimate produced by molecular docking to rank the predicted favorability of a ligand's binding pose or interaction with a target.

A docking score is a numerical estimate of how favorably a small molecule, or ligand, binds to a target such as a protein. Molecular docking explores possible binding poses and scores them to rank candidates for further study. The score is an approximation, not a measurement of actual binding affinity. section 9.0 section 9.1.5

In a Vina docking result, the affinity column gives the score for each pose in kcal/mol-like units. Lower scores indicate more favorable poses. Other columns describe interaction energy and internal strain; they are useful diagnostics, but poses are ranked by affinity alone. section 9.1.5

Comparing poses

A nine-pose result ranks pose 0 first at −9.246, followed by pose 1 at −8.543. The difference is about 0.703 kcal/mol. A better-ranked pose is a candidate for inspection, not proof that the ligand binds as predicted. section 9.1.5

# Output of print(results) —9 poses, sorted best to worst:
   affinity   inter  intra  torsions  intra_best_post
0    -9.246 -13.571 -1.112     4.324           -1.112
1    -8.543 -11.782 -1.868     3.995           -1.112
2    -8.531 -12.003 -1.629     3.990           -1.112
3    -8.480 -11.860 -1.698     3.966           -1.112
4    -8.353 -12.081 -1.291     3.907           -1.112
5    -8.298 -11.132 -2.159     3.881           -1.112
6    -8.084 -11.139 -1.837     3.781           -1.112
7    -7.958 -11.056 -1.736     3.722           -1.112
8    -7.868 -10.555 -2.104     3.680           -1.112

Compare the affinity ranking with the diagnostic energy columns across all nine poses. section 9.1.5

Check promising scores before drawing conclusions:

  • Inspect top poses for favorable contacts, steric clashes, and realistic conformations. section 9.1.5
  • Compare with the original ligand pose as a control, or rescore top poses with other scoring functions and look for agreement. section 9.1.5
  • Use molecular dynamics to assess pose stability or experiments to determine actual binding affinity. Published score thresholds are not comprehensive or precise. section 9.1.5

For very large libraries, docking every molecule can be expensive. Progressive docking uses scores from an initially docked subset to train a surrogate model that predicts score proxies for other molecules. An active learning loop selects another batch, obtains benchmark scores from a docking oracle, and retrains the model within a computational budget. section 9.0 section 9.2.7

Tracking screening progress

In a 4AG8 dataset experiment, a proxy oracle looks up previously computed scores rather than running docking for each test. After 7,500 molecules had been selected, the run had found 73 of the dataset’s top 100 molecules by docking score. This measures recovery against that reference ranking, not experimental binding success. section 9.2.6 section 9.2.7

# Output of results_df.head(10).to_string() — per-iteration progress:
   iteration  total_molecules  top_molecules_found  best_score        time
0          1             5250                   13      -20.57        2.60
1          2             5500                   19      -20.57        4.86
2          3             5750                   32      -20.66        1.84
3          4             6000                   41      -20.66        1.22
4          5             6250                   51      -20.66        1.23
5          6             6500                   58      -20.66        1.32
6          7             6750                   61      -20.66        1.89
7          8             7000                   67      -20.66        1.78
8          9             7250                   71      -20.66        1.47
9         10             7500                   73      -20.66        1.49

See how the best score and number of top-reference molecules found change over ten iterations. section 9.2.7

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Explains pose ranking, diagnostic energy terms, and validation options, without detailing the scoring-function formula.
Examples
Compares nine Vina poses and tracks recovery of the top 100 scored molecules in a 4AG8 dataset.
What you can do after
Evaluate docking poses and screening progress using affinity rankings, diagnostic terms, validation checks, and reference-set recovery.
How it connects
Connects molecular docking to surrogate models, active learning, molecular dynamics, and experimental affinity validation.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Build AI Drug Discovery Pipelines. The section links point to the parts of the book it draws on.