Fig. 7

Development of a multi-chain AIRR predictor for rheumatoid arthritis. a Predictive performance of the optimal classifier that aggregates chain and gene segment usage information. On the left side, the graphical representation of the confusion matrix is shown. The number of individuals that were predicted as patients or controls is shown between parenthesis. On the right side, the value of the five metrics used for evaluating the predictive performance is shown. Abbreviations: CTRL, controls; RA, rheumatoid arthritis; FN, false negatives; FP, false positives; FPR, false positive rate; TP, true positives; TN, true negatives. b Receiver operating characteristic and precision-recall curves illustrating the diagnostic ability of the multi-chain AIRR classifier for rheumatoid arthritis. Abbreviations: ROC, receiver operating characteristic; ROCAUC, are under the receiver operating characteristic curve; PRAUC, are under the precision-recall curve. c Identification of the 15 AIRR predictor variables that contributed most to the diagnosis of patients with rheumatoid arthritis in accordance with the mean decrease in accuracy and Gini importance measures