Skip to main content
. 2020 Aug 4;27(8):1190–1203. doi: 10.1089/cmb.2019.0337

FIG. 4.

FIG. 4.

Tuning the deep neural network architecture using grid search. We varied the number of latent units (16, 32, 64), together with three different configurations for the number of hidden layers (1, 2, 3) and the number of hidden units (32, 128, 512). For evaluation, we use combined correlation scores of Pearson and Spearman's correlation calculated using all marker gene and protein pairs on the test set. Darker blue color represents better performance.