liveBook · Manning

Q2

As explained in Build AI Drug Discovery Pipelines

Q2 is a QSAR predictive-performance metric analogous to R2 but calculated for held-out test data.

Q2 measures QSAR predictive performance on held-out test data. Like R2, it summarizes model fit, but Q2 specifically evaluates predictions for unseen compounds rather than training observations. section 6.3.1

Use Q2 when comparing candidate QSAR models:

  • Screen candidates for Q2 > 0.7 alongside training-data fit of R2 > 0.7. section 6.3.1
  • Prefer the model with the highest Q2 after checking descriptor significance; resolve ties with the simplest model. section 6.3.1
  • Interpret a single-split Q2 cautiously because small datasets, split choices, feature filtering, and test-set leakage can substantially change it. section 6.3.1

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Defines Q2 by contrast with R2 and applies a threshold, but does not provide its mathematical formula.
Examples
Uses candidate QSAR models for small-molecule RNA-target ligands and evaluates them on a selected held-out test set.
What it assumes
Assumes familiarity with QSAR models, descriptors, train-test splits, and R2; distinguishes predictive testing from training fit.
What you can do after
Evaluate QSAR generalization by comparing predictions for unseen compounds and ranking models by Q2.
How it connects
Connects Q2 with QSAR modeling, R2, multiple linear regression, descriptor significance, model selection, and Occam’s razor.

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.