Fig. 4
From: Predicting chemotherapy response using a variational autoencoder approach
Comparing the distributions of feature importance scores for the most informative features for PCA-XGBoost and VAE-XGBoost, for predicting response to chemotherapy in colon cancer (COAD). Bars indicate the sum (over 30 replications) of XGBoost feature importance scores. “Group” indicates the low-dimensional embedding method used (VAE or PCA). Bars separately ordered from highest to lowest (only top 20 most important features shown)