Data bias is systematic distortion in a dataset that causes some conditions, entities, or populations to be overrepresented or underrepresented.
Training data bias is systematic overrepresentation or underrepresentation that limits where a model’s predictions generalize. In drug discovery, it can reflect experimental practices rather than the biological or chemical population of interest. section 13.4.3
Common consequences include:
Models trained on heterogeneous measurements inherit uncertainty from different assay protocols, instruments, reagent lots, and experimental conditions. section 13.4.3
AlphaFold2’s training data contains 91% soluble proteins but only 3–4% membrane proteins, reducing accuracy for membrane targets such as GPCRs and ion channels. section 13.4.3
Small-molecule-heavy datasets underrepresent macrocycles and other larger therapeutic classes, while clinical datasets underrepresent racial, ethnic, geographic, and age groups. section 13.4.3
How Build AI Drug Discovery Pipelines covers it
How deep it goes
Defines representation bias and illustrates its effects with protein, chemical, assay, and clinical-population examples.
Examples
Uses protein databases, small-molecule datasets, assay measurements, and historically unbalanced clinical-trial populations as examples.
What you can do after
Evaluate model applicability by comparing its training distribution with the intended proteins, molecules, measurements, or patient populations.
How it connects
Connects training data bias to measurement noise, generalization, ADMET prediction, AlphaFold2, and health disparities.
Grokking StatisticsData bias is systematic distortion in collected or represented data that causes analyses or models to produce unfair or unrepresentative results.
biaskind of · Bias is a systematic tendency or deviation that distorts measurements, estimates, judgments, or outcomes relative to a relevant reference or target.
fusionproduced by · Fusion is the combination of information, signals, or results from multiple sources into a unified representation, result, or inference.
A/B testcontrasts with · An A/B test is a controlled experiment that compares outcomes for two variants assigned to different users or experimental units.
machine learningaddresses · Machine learning is a field of computational methods in which systems learn patterns or relationships from data to make predictions, decisions, or other outputs.
data collectionproduced by · Data collection is the systematic process of acquiring observations, measurements, records, or other information from one or more sources for a defined purpose.
language modelconfigures · A language model is a computational model that represents patterns in linguistic sequences to assign probabilities, predict, generate, or evaluate language.
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.