liveBook · Manning

Data quality

As explained in Build AI Drug Discovery Pipelines

Data quality is the degree to which data is accurate, consistent, complete, reliable, and fit for its intended analytical purpose.

Data quality is the reliability and suitability of data for drug-discovery models. Poor measurements, inconsistent protocols, and biased datasets can limit predictions even when algorithms are sound. section 13.4.3

Important failure sources include:

  • Different laboratories can measure the same molecule's ADMET properties differently because of assay protocols, instruments, reagent lots, and experimental conditions. section 13.4.3
  • Training on heterogeneous measurements introduces uncertainty from experimental noise rather than from the model's algorithm. section 13.4.3
  • Underrepresented membrane proteins, larger molecules, macrocycles, and diverse patient populations can restrict how well predictions generalize. section 13.4.3

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Explains measurement noise and representation bias with concrete cases, but does not provide a remediation workflow.
Examples
Uses cross-laboratory ADMET variation, protein-database composition, chemical-class imbalance, and demographic underrepresentation as examples.
What you can do after
Evaluate model reliability by distinguishing algorithmic limitations from measurement noise and training-data underrepresentation.
How it connects
Links data quality to ADMET prediction, AlphaFold2, membrane proteins, Lipinski's Rule of Five, and clinical prediction models.

Also explained in

Compare how 15 books explain data quality →

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.