Skip to main content

Table 4 Comparison of chemotherapy response prediction performance (AUPRC) for XGBoost models trained with original transcriptome data (“Raw data”) or transcriptome data encoded with PCA, ICA, or VAE. This analysis was carried out across five cancers (BRCA, COAD, BLCA, PAAD, and SARC)

From: Predicting chemotherapy response using a variational autoencoder approach

  AUPRC (mean) p (Welch’s t-test)
Cancer type VAE Raw data PCA ICA VAE versus Raw data VAE versus PCA VAE versus ICA
BRCA 0.192 0.157 0.145 0.150 \(4.21 \times 10^{-6}\) \(7.42 \times 10^{-10}\) \(5.01 \times 10^{-8}\)
PAAD 0.764 0.729 0.746 0.713 \(3.38 \times 10^{-4}\) \(9.12 \times 10^{-2}\) \(2.02 \times 10^{-6}\)
COAD 0.593 0.535 0.579 0.545 \(6.52 \times 10^{-4}\) \(3.91 \times 10^{-1}\) \(1.27 \times 10^{-3}\)
BLCA 0.649 0.623 0.587 0.654 \(4.30 \times 10^{-2}\) \(1.60 \times 10^{-7}\) \(6.13 \times 10^{-1}\)
SARC 0.736 0.713 0.714 0.729 \(6.15 \times 10^{-2}\) \(1.61 \times 10^{-1}\) \(5.96 \times 10^{-1}\)
  1. The p values are for row-wise difference of means tests for the indicated pairs of sample groups (columns). For each cancer type (row), the highest mean AUPRC performance is shown in boldface