Skip to main content
Fig. 3 | BMC Medical Genomics

Fig. 3

From: Selecting precise reference normal tissue samples for cancer research using a deep learning approach

Fig. 3

Applying an autoencoder for representing gene expression profiles. a Schema and parameters. Both encoder and decoder have one layer in addition to the input/output layer. The input of encoder and the output of decoder are the expression of 60,498 transcripts. The objective function is to minimize the difference between the output and input. Sixty-four encoded features are used to represent expression profiles. Between layers, the following functions Leaky ReLU activation, batch normalization, and drop out are applied. Both network architecture and parameters can be changed. b MSE loss for the training and test set. Lower MSE loss means the output is more similar to the input. c t-SNE distribution of all samples using encoded features from an autoencoder. Dots were colored by data resources

Back to article page