liveBook · Manning

Williams plot

As explained in Build AI Drug Discovery Pipelines

A Williams plot is a diagnostic plot relating observation leverage to standardized residuals for assessing outliers and model applicability.

A Williams plot diagnoses QSAR model applicability by relating each observation’s leverage to its standardized residual. Low leverage and small residuals indicate points within the model’s applicability domain. section 6.3.1

Interpret the plot using two types of warning boundary:

  • Flag large standardized residuals as observations whose prediction errors are unusually large. section 6.3.1
  • Flag high leverage as observations exerting unusual influence on the model. section 6.3.1
  • Use standardized-residual bounds of +3 and −3, and a leverage bound of 3(p + 1)/n. section 6.3.1

QSAR binding model

For an lnKoff model, the plot identifies one training sample outside the applicability domain: the ligand DMZ-p8. section 6.3.1

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Explains leverage, standardized residuals, outlier interpretation, and threshold formulas, but does not detail plot construction.
Examples
Uses an lnKoff QSAR plot that identifies the ligand DMZ-p8 as outside the applicability domain.
What it assumes
Assumes a fitted QSAR model with observed and predicted values, descriptors, and data-point counts.
What you can do after
Diagnose model outliers and applicability-domain violations using leverage and standardized residuals.
How it connects
Connects QSAR modeling, applicability domain, leverage, standardized residuals, and outlier detection.

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.