Additional file 34 of Seven-chain adaptive immune receptor repertoire analysis in rheumatoid arthritis reveals novel features associated with disease and clinically relevant phenotypes

Additional file 34: Table S27. Predictive performance of the multi-chain AIRR predictors evaluated in the present study. The predictive performance of each multi-chain AIRR predictor is determined using the accuracy, sensitivity, specificity, precision and false positive rate measures. For each multi-chain AIRR predictor, the number of individuals that were properly classified is also shown. The AIRR feature column indicates if the multi-chain AIRR predictor was built using information from a single AIRR- feature or aggregating information from different AIRR features. Information from different AIRR features was aggregated according to different thresholds of accuracy when the sample size of the classes tended to be balanced (i.e., predictor of rheumatoid arthritis diagnosis and disease activity). The aggregation of information from different AIRR features was based on different specificity thresholds when the clinical interest was to identify the minor class (i.e., predictors of TNFi response, ACPA and RF phenotypes). The predictive performance of redundant multi-chain AIRR predictors obtained after applying different accuracy or specificity thresholds is not shown.