Table 2 Comparison of chemotherapy response prediction performance for XGBoost models trained with VAE-derived features versus autoencoder (AE)-derived features, for three cancer types (BRCA, BLCA, and PAAD)
From: Predicting chemotherapy response using a variational autoencoder approach
Cancer type | Mean | p (Welch’s t-test) | ||||
|---|---|---|---|---|---|---|
AUROC | AUPRC | AUROC | AUPRC | |||
VAE | AE | VAE | AE | VAE versus AE | VAE versus AE | |
BRCA | 0.674 | 0.575 | 0.192 | 0.137 | \(1.61 \times 10^{-15}\) | \(5.38 \times 10^{-10}\) |
PAAD | 0.738 | 0.660 | 0.764 | 0.695 | \(3.46 \times 10^{-10}\) | \(6.72 \times 10^{-7}\) |
BLCA | 0.659 | 0.573 | 0.649 | 0.577 | \(7.97 \times 10^{-12}\) | \(1.23 \times 10^{-7}\) |
SARC | 0.704 | 0.611 | 0.736 | 0.654 | \(2.78 \times 10^{-7}\) | \(1.75 \times 10^{-6}\) |