GSAE: an autoencoder with embedded gene-set nodes for genomics functional characterization

BMC Systems Biology

Table 1 Evaluation of the clustering performance of the two t-SNE results in Fig. 2. As a reference, the compression rate from 15,975 features down to 200 supersets is about 98.7%

Index Method	t-SNE of 15,975 genes	t-SNE of 200 supersets	Compression loss^a
Dunn index	0.247	0.189	23.48%
Silouette index	0.355	0.358	−0.85%
IID index	7.924	8.125	−2.54%

^aThe compression loss = (index score of genes – index score of supersets) / index score of genes

ISSN: 1752-0509