Fig. 4
From: Gene signature for the prediction of the trajectories of sepsis-induced acute kidney injury
Development of a support vector machine model using genetic algorithms. A Stability of gene ranks over the 1000 evolution cycles. The plot shows the stability of the rank of the top 50 genes, which is designed to aid in the decision to stop or continue the process once the top ranked genes are stabilized. When genes have many changes in ranks, the plot show different colours; hence the rank of these genes is unstable. Commonly the top 2 “black” genes are stabilized quickly, in 50 to 200 solutions (evolutions), whereas low ranked “grey” genes would require many thousands of solutions to be stabilized. B heatmap plot showing the scaled gene expression abundance grouped by AKI groups. The genes displayed were selected by classical forward selection method, adding one gene at the time starting from the most frequent to the least frequent. C Hyperparameter tuning for training the SVM for the gene model. Contour plot shows the hyperparameter tuning process by the grid search method. Cost and gamma are two hyperparameters of the SVM model. The plot shows the accuracy of the SVM model (denoted by color) at each combination of cost (vertical axis) and gamma (horizontal axis), and the combination of the hyperparameters at the highest accuracy is used to train the final model. D Comparisons of the SVM models based on clinical variable and gene features. The gene model outperformed clinical model as indicated by significantly higher values of AUC