Fig. 1.

(a) Diagram of CombNet. Microarray and RNA-seq data were mixed before entering the autoencoder. Same feature spaces were defined in both platforms (b) Diagram of CrossNet. The first part (generative part) is an autoencoder, where an encoder and decoder are combined to regenerate microarray gene expression profile. The second part (discriminative part) is then introduced to reduce the difference between regenerated microarray data (i.e., the output of generative part) and origin RNA-seq data. In current version, we do not build another discriminative model but use the crossentropy to simplify the process