Fig. 4From: Pharmacovariome scanning using whole pharmacogene resequencing coupled with deep computational analysis and machine learning for clinical pharmacogenomicsThe comparison of average accuracy for the designed ML model by means of both internal and external cross-validation. Accuracy of the prediction model for developing ADRs in cardiovascular patients demonstrated by different cross-validation approaches. A The RFE process of machine learning reduced the variables to 175 important genotype variants. These are the final variants indicated by the RF model, which employed internal fivefold cross-validation. B The accuracy changed during the testing of the designed model by external cross-validation and the number of important pharmacovariants reduced to 60. The subset of the variants achieved an average accuracy of 0.9818 and 0.9512 on predicting whether a patient will have ADRs or not, respectively. ML: machine learning, ADR: adverse drug reactions, RFE: recursive feature elimination, RF: random forestBack to article page