Sensitivity and specificity for random forest tests applied to peptide-MHC binding scores for vaccine classification of Benchmark dataset.

Abbreviations: (R) = target variable e.g. 1 or 0 in training data randomly changed for each protein, HE = hold-out dataset error (%) i.e. error when predicting 30% of training data, OE = overall error (%) i.e. percentage of incorrect predictions, SN = sensitivity (%) = true positives/(true positives+false negatives), SP = specificity (%) = true negatives/(true negatives+false positives). aCross-validation involved a random sample of 70% from training dataset to build predictive model and remaining 30% used for testing. This was repeated 10 times and predictions averaged (predictions for the same input data fluctuate unless a random seed is set initially). bBenchmark are proteins from published studies with known or expected T-cell responses (source species: T. gondii) –100% from training data used to build predictive model. Note: Number of input variables used to build predictive model = 304 (i.e. number of allele-peptide length combinations derived from 76 common alleles). Sensitivity and specificity for random forest tests applied to peptide-MHC binding scores for vaccine classification of Benchmark dataset.