Skip to main content

Table 1 Disentanglement metrics for two splatter-simulated single-cell RNA-seq datasets with four ground truth variables

From: MichiGAN: sampling from disentangled representations of single-cell data using generative adversarial networks

   Spearman correlation gap FactorVAE metric MIG
Linear step PCA 0.68 ±0.00 0.35 ±0.01 0.54 ±0.00
  VAE 0.3 ±0.04 0.4 ±0.02 0.48 ±0.13
  β-TCVAE 0.18 ±0.05 0.48 ±0.03 0.72 ±0.02
Non-linear step PCA 0.72 ±0.00 0.35 ±0.01 0.55 ±0.00
  VAE 0.27 ±0.07 0.41 ±0.02 0.43 ±0.08
  β-TCVAE 0.16 ±0.06 0.51 ±0.04 0.66 ±0.16
  1. The mean and standard deviation over 5 runs are presented for each method. The dimensionality of the latent space was 10 for all three approaches