Figure 1.

Study design

(A) Demonstration of the training data given by the DREAM Challenge.

(B) Strategy of training in vivo malaria ART prediction models and transferring the model to in vitro malaria transcriptome datasets. First, we imputed missing values and rank-normalized the expression data. Second, we cross-validated models of different base learners. We then selected the base learner and sample conditions with best performance by cross-validations and reverse test. Lastly, important predictive biomarkers are prioritized by SHAP analysis.