Fig. 9

Visualization of a simple example of probabilistic random walks demonstrating the importance of walk number to avoid overfitting when training a model with gene co-expression networks. A small number of walks is represented by the image on the left, where the blue arrows indicate a walk that is going through a hub gene (center node) in the network. A larger number of walks through this same node is shown on the right. Each arrow colour indicates a separate walk. Assuming that all of these edges have comparable edge weights, if we have a small number of walks that travel through this node, which is more likely the fewer/shorter walks we make, than this gene will not be well represented in the final model, no matter how many iterations/epochs are made during the training. This challenge would become even greater when hub genes are also connected to each other which increases the possibilities for distinct walks immensely. Densely connected nodes tend to be a characteristic of gene co-expression networks, so the number of walks is an important consideration