Fig. 5From: Pharmacovariome scanning using whole pharmacogene resequencing coupled with deep computational analysis and machine learning for clinical pharmacogenomicsComparison of ML final selected variants’ importance. 175 final variants by internal and 60 variants by external cross-validation, introduced by the RF model as the important variables that may cause ADRs in cardiovascular patients who received particular drugs. The differences and low accuracy for external validation must be considered while applying machine learning for small cohorts (see the text for the discussion). Associated importance values are available in Additional file 3. ML: machine learning, ADR: adverse drug reactions, RF: random forestBack to article page