Additional file 1 of Calibrating multiplex serology for Helicobacter pylori

Additional file 1: Table S1: Average out-of-bagroot mean square error for different hyperparameter combinations for the BART algorithm on the training data. Scores are displayed for all hyperparameter combinations in order of increasing RMSE. Averages are over validation folds within a tenfold cross-validation scheme. The hyperparameters $$m$$ m and $$k$$ k represent the number of trees and the amount of prior information used in the model, respectively. The best hyperparameter combination, $$\{k=2, m=200\}$$ { k = 2 , m = 200 } , was used for prediction among cases. Table S2: Average accuracy scores and standard deviations (SDs) for different hyperparameter combinations for the mBART algorithm on the training data. Averages and SDs are over validation folds within a tenfold cross-validation scheme. Scores are displayed for all hyperparameter combinations in order of decreasing accuracy. The hyperparameters $$m$$ m and $$k$$ k represent the number of trees and the amount of prior information used in the model, respectively. The best hyperparameter combination (that which yielded the optimal score across the 10 folds) was used for prediction in the case data. Due to rounding, the accuracy scores of the combinations $$\left\{k=2, m=200\right\}$$ k = 2 , m = 200 and $$\{k=3, m=200\}$$ { k = 3 , m = 200 } appear equal; however, the second set has a higher average score and a lower SD. Table S3: Average accuracy scores and standard deviations (SDs) for different hyperparameter combinations for the random forest algorithm on the training data. Averages and SDs are over validation folds within a tenfold cross-validation scheme. Scores are displayed for all possible hyperparameter values in order of decreasing accuracy. The hyperparameter $${m}_{try}$$ m try represents the number of variables used for node splitting. The best hyperparameter combination (that which yielded the optimal score across the 10 folds) $$\{{m}_{try}=6\}$$ { m try = 6 } , was used for prediction in the case data. Figure S1: Misclassification error for different values of $$\lambda$$ λ in the elastic net (EN) algorithm on the training data within a tenfold cross-validation (CV) scheme. Red dots and grey bars represent the average misclassification error and standard deviation of misclassification errors, respectively, across the 10 CV folds for each value of $$\lambda$$ λ . The dashed vertical lines are plotted at the value of $$\lambda$$ λ with the minimum average CV error (first dashed line from the left) and at the largest value of $$\lambda$$ λ such that the error is within one standard error of the minimum (second dashed line from the left). The numbers displayed at the top of the plot represent the number of non-zero coefficient estimates at the corresponding values of $$\lambda$$ λ . The value of $$\lambda$$ λ with the minimum average CV error $$(log(\lambda)= -3.93)$$ ( l o g ( λ ) = - 3.93 ) was used within the EN model for prediction on the case data. The value of $$\alpha$$ α (EN mixing parameter) was fixed at 0.90 to allow more weight towards a lasso penalty ( $$\alpha$$ α = 1) while still allowing for a ridge penalty ( $$\alpha=0$$ α = 0 ) to control multicollinearity effects.