liveBook · Manning

Similarity map

As explained in Build AI Drug Discovery Pipelines

A similarity map is a molecular visualization that assigns atom-level weights according to how much each atom affects a model prediction when its contribution is ablated.

A similarity map visualizes which parts of a molecule influence a model’s prediction by assigning weights to individual atoms or substructures. In drug-discovery classification, it makes Morgan-fingerprint reasoning visible on the molecular structure. section 5.3.2

A practical workflow uses:

  • A rendered molecular image whose highlighted regions indicate contributions to CYP 3A4 inhibition classification. section 5.3.2

CYP 3A4 inhibition

Apply the map to a validation-set molecule and a fitted Random Forest classifier to inspect which molecular regions support its predicted inhibitor probability.

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Demonstrates an RDKit implementation that maps Morgan-fingerprint model probabilities onto atoms, without deriving the mapping internals.
Examples
Uses a validation-set molecule, a CYP 3A4 classifier, and a rendered RDKit similarity-map figure.
What it assumes
Assumes a fitted molecular classifier and Morgan fingerprints while providing implementation wrappers for model predictions.
What you can do after
Apply it via RDKit to visualize which molecular regions support a model’s class-probability estimate.
How it connects
Connects Random Forests, Morgan fingerprints, feature importance, molecular substructures, and CYP 3A4 inhibition.

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.