if your feat is the one hot encode with dim of (x size that not 1) with node (N) you may add the another dimension for graph the final shape will be torch.size(1, N_node, M_feat). as you may see in the example of "adjacency matrix" Do you work on drug - ligand interaction?