liveBook · Manning

Quality

As explained in Build AI Drug Discovery Pipelines

A scoring function is a function that assigns a numerical value to a candidate according to specified criteria.

A scoring function assigns a numerical value to each candidate according to specified criteria. In drug discovery, it turns predicted activity, molecular similarity, docking energy, or other evidence into a value that supports ranking and prioritization. section 8.1.3 section 9.1.2

Virtual screening uses scores to place likely active compounds near the top of a ranked list, allowing a small number of candidates to be selected before testing the entire library. Similarity searches, neural networks, linear regressors, and molecular docking can all provide scoring functions. section 8.1.3

EF(X%)=(Number of actives found in top X% of the ranked list)Total actives in dataset×(X%100)

Enrichment factor evaluates whether a scoring function finds more active compounds in the top fraction of a ranked list than random selection would find. section 8.1.3

Docking scores rank alternative ligand poses by estimated favorability. The Vina scoring function combines interaction-related terms such as van der Waals interactions, electrostatics, hydrogen bonding, desolvation, internal ligand energy, and torsional effects. Lower affinity scores indicate more favorable predicted poses, but visual inspection and additional validation remain necessary. section 9.1.2 section 9.1.5

EGFR virtual screening

A neural network predicts EGFR pIC50 values for compounds in the Hinge Binders Library. After removing compounds already used for training, the workflow ranks candidates by predicted activity and retains the top 100 for experimental assessment. section 8.1.3

Docking pose selection

For one ligand, nine poses are sorted by Vina affinity; the best pose has an affinity of -9.246 kcal/mol-like units. Additional interaction, intramolecular, and torsion columns diagnose strain, while affinity alone determines the ranking. section 9.1.5

A score is only as useful as the criteria and validation behind it. Weak property-prediction scores can cause generative models to exploit defects, proposing molecules that score well but are unstable, unsynthesizable, or unrealistic. section 1.3.2

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Explains scoring intuitively and operationally, including ranking, enrichment evaluation, docking terms, diagnostics, and validation limits.
Examples
Uses EGFR pIC50 ranking, Hinge Binders Library prioritization, enrichment factors, and Vina docking affinities for concrete examples.
What you can do after
Evaluate candidates by ranking scores, calculate enrichment factors, inspect docking diagnostics, and recognize when scores require further validation.
How it connects
Connects scoring functions to virtual screening, supervised learning, molecular docking, pose evaluation, generative models, and reward hacking.

Also explained in

Compare how 10 books explain quality →

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.