Figure 2.

Prediction of AML in Random and Cross-study Sampling Scenarios

(A) Schema illustrating the approach to predict AML in random and cross-study sampling scenarios.

(B–D) AML classification accuracies based on the lasso model of AML versus all other samples and for both sampling strategies are shown for dataset 1 (B), dataset 2 (C), and dataset 3 (D).

(E and F) Classification accuracies for the differential diagnosis case (AML versus other leukemic samples, namely, AML, ALL, CML, CLL, and MDS) for both sampling strategies are shown for dataset 1 (E) and dataset 2 (F). Mean accuracies of the lasso models are shown as a function of the training sample size n_train. Results are over 100 random training and test sets, with error bars indicating the standard deviation.

(G) Comparison of the performance of the LASSO models introduced in panels A to F with a neural network approach using either 5 or 10 layers. Error bars indicate the standard deviation.

See also Figures S3–S8 and S13, and Tables S2 and S4.