Skip to main content
. 2022 Feb 23;8:9. doi: 10.1038/s41540-022-00218-9

Fig. 1. Design of the artificial deep neural networks.

Fig. 1

The networks were made to predict the expression of 25,861 genes based on 1,625 gene regulators, i.e. transcription factors (TFs). We used more than 100,000 randomly drawn RNA-seq samples from the ARCHS4 database to train the models. Moreover, we designed 15 DNNs of 1–3 hidden layers, and one shallow neural network without any hidden layers. For the DNNs, each hidden layer consisted of either 50, 100, 250, 500, or 1000 hidden nodes using the exponential linear unit, ELU as activation function.