liveBook · Manning

Diversity-based sampling

As explained in Build AI Drug Discovery Pipelines

Diversity selection chooses candidates to maximize coverage or variation across the represented candidate space.

Diversity-based sampling selects unlabeled candidates that are maximally different from already labeled examples, using distances in feature space to broaden coverage. section 9.2.5

In virtual screening, it explores distinct chemical scaffolds instead of repeatedly testing close variants from already-sampled molecular families, increasing the chance of finding novel actives. section 9.2.5

Common representativeness strategies include:

  • Select maximally different candidates with feature-space distance metrics. section 9.2.5
  • Combine uncertainty with density estimation to prioritize uncertain candidates in high-density regions. section 9.2.5
  • Partition the unlabeled pool into clusters and choose representatives from each cluster. section 9.2.5

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Defines distance-, density-, and clustering-based strategies but does not provide a dedicated diversity implementation.
Examples
Uses diverse chemical scaffolds, labeled examples, feature-space distances, density-weighted candidates, and clustered unlabeled pools.
What it assumes
Assumes labeled and unlabeled examples represented in an input or feature space.
What you can do after
Apply it to select batches that cover varied chemical scaffolds and reduce over-specialization in sampled regions.
How it connects
Connects representativeness with active learning, uncertainty sampling, density estimation, clustering, and exploration-exploitation.

Also explained in

Compare how 2 books explain diversity-based sampling →

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.