Fig. 6

Classifiers for ALS diagnosis. a Random forest variable importance scores. Scores are shown for the top 30 genes. Tuning parameters yielding the highest out-of-bag (OOB) accuracy in preliminary trials were used (450 input genes, ntree = 400, mtry = 50). b Random forest prediction accuracy. The expression of 450 input genes was used as predictors. c Random forest PC analysis parameters. d Random forest prediction accuracy (with PC score predictors). e Logistic regression PC analysis parameters. f Logistic regression accuracy (PC predictors). g SVM PC analysis parameters. h SVM cost and gamma parameters. i SVM prediction accuracy. In b, f and i, histograms show the accuracy obtained across cross-validation trials (upper right: proportion of trials with accuracy significantly greater than non-information rate (NIR) of 50%, i.e., McNemar’s test). For each cross-validation trial, 592 subjects were used for training (296 ALS patients vs. 296 CTL/MIM subjects) and 200 subjects were used for testing (100 ALS patients vs. 100 CTL/MIM subjects). In c, e, g and h, cross-validation accuracy is shown for the analysis parameters as indicated on vertical and horizontal axes. For c, e and g, the number of PCs evaluated for each square is equal to the number of input genes (left axis) multiplied by the percentage of PCs (bottom axis). The 10 parameter combinations with highest accuracy are labeled (1 = highest accuracy). j Paired sensitivity and specificity estimates from 72 prior studies. Studies with disease control cohorts (diamonds) used patients with non-ALS neurological diseases, or the combination of healthy controls and patients with non-ALS neurological diseases. Dashed brown lines denote sensitivity and specificity estimates from the current study. k Min(Sens, Spec) versus sample size. For each pair of sensitivity and specificity estimates, the lower value is plotted (vertical axis) relative to the lower of the two ALS and CTL cohort sample sizes (horizontal axis). Dashed brown lines denote values from the current study