Fig. 4From: Baseline antibody profiles predict toxicity in melanoma patients treated with immune checkpoint inhibitorsDevelopment of classification models to predict immunotherapy toxicity using antibodies from pre-treatment melanoma patient sera. a Scatterplot showing distribution of decision values from support vector machine (SVM) classifier models based on “filtered” antibody (feature) lists for prediction of severe toxicity. Data summarizes training and testing results from 100 repetitions of fivefold cross validation for pre-anti-CTLA-4 samples. Gold circles represent true positives (severe toxicity sample called as severe toxicity) and green crosses represent true negatives (no/mild toxicity sample called as no/mild toxicity). Red circles represent false negatives (severe toxicity sample called as no/mild toxicity) and blue crosses represent false positives (no/mild toxicity called as severe toxicity). b As for a, but summarizing 100 repetitions of fivefold cross validation for anti-PD-1 samples. c As for a, but summarizing 100 repetitions of threefold cross validation for anti-CTLA-4 and anti-PD-1 combination samples. d Summary of accuracy, sensitivity, and specificity cross validation statistics based on SVM models for prediction of toxicity in anti-CTLA-4 samples (no/mild toxicity, n = 30; severe, n = 9). e As for d, but for anti-PD-1 samples (no/mild toxicity, n = 19; severe, n = 9). f As for d, but for combined anti-CTLA-4 and anti-PD-1 samples (mild toxicity, n = 4; severe, n = 7)Back to article page