Train a GCN without any feature?

I was using Graph convolutional network (GCN) in PyTorch to train a graph of nodes that has a couple of features, but I’m wondering if I have some of the nodes with no features (these nodes are kind of knowledge/metadata); can I still do the training on the graph including these nodes?

If not, what would be an alternative solution to keep training on them?

Any ideas?

Any suggestion or idea if that’s possible?