Fig. 3: Effects of data downsampling and augmentation on classification and clustering. | Nature Communications

Fig. 3: Effects of data downsampling and augmentation on classification and clustering.

From: Realistic in silico generation and augmentation of single-cell RNA-seq data using generative adversarial networks

Fig. 3

a F1 score of an RF classifier trained to discriminate cluster 2 test from other test cells when trained on training (red), upsampled (orange), or augmented (blue) cells for eight levels of downsampling (50% to 0.5%). bd t-SNE representation of cluster 2 real test (red) and cscGAN-generated (blue) cells for three levels of downsampling (10% panel b, 2% panel c, and 0.5% panel d). Other test cells are shown in gray.

Back to article page