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

BMC Systems Biology

Table 4 The size of encoder layers and the 10-fold cross-validation accuracy of each neural network classifier

NN classifier^a	Input Type	Encoder Layer 1^d	Encoder Layer 2	Encoder Layer 3	Encoder Layer 4	Accuracy of 10-fold cross validation
Superset	Genes^b	2334	200			88.79%
Gene set	Genes	2334				87.69%
2-layer fc	Genes	2334	200			47.86%
2-layer fc	Genes	2000	500			37.98%
4-layer fc	Genes	2000	200	100	50	46.06%
2-layer fc	PC^c	400	100			87.57%
4-layer fc	PC	200	200	100	25	87.57%

^a2-, 4-layer fc: 2- or 4- layer fully connected AE
^bGenes input is the 15,183 genes of TCGA BRCA RNA-seq data
^cPC input is the top 500 principal components of PCA analysis
^dThe encoder layer 1 of superset and gene set classifier is the gene set layer (not a fully connected layer)

ISSN: 1752-0509