liveBook · Manning

Odds ratio

As explained in Build AI Drug Discovery Pipelines

An odds ratio is the ratio of two odds, often comparing the odds under a changed feature value with the odds under a reference value.

An odds ratio compares two odds: the odds of an outcome after increasing one feature by one unit versus the odds with that feature unchanged. In logistic regression, it translates a model weight into a practical change in predicted odds. section 5.1.3

Interpret an odds ratio by holding every other feature fixed:

  • A positive weight increases the log-odds and produces an odds ratio greater than one. section 5.1.3
  • A negative weight decreases the predicted odds; the change describes model output, not a guaranteed biological effect. section 5.1.3

CYP3A4 inhibition

Suppose a molecular feature has weight 0.45 in a classifier predicting whether a molecule inhibits CYP3A4. Increasing that feature by one unit gives an odds ratio of exp(0.45) = 1.57, or a 57% increase in predicted inhibitor odds, with other features unchanged. section 5.1.3

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Derives the one-unit feature relationship and works through a numerical example, while stopping short of broader causal interpretation.
Examples
Uses a molecular feature with weight 0.45 to demonstrate a 1.57 odds ratio and 57% predicted increase.
What it assumes
Assumes familiarity with probability, logistic regression, sigmoid outputs, and log-odds.
What you can do after
Evaluate model weights as one-unit changes in predicted outcome odds while holding other features fixed.
How it connects
Connects odds ratios to logistic regression, sigmoid functions, log-odds, molecular features, and CYP3A4 inhibition.

Also explained in

Compare how 3 books explain odds ratio →

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.