Modifying graph convolution layer (message passing)

Hi, I am new to pytorch geometrics. I want to make a change to a graph convolution layer message passing scheme. more specifically I want to add some equation to the massage passing so I can remove some nodes from aggregation. I have been trying to figure out that what part of this code I can modify to achieve this .the following is the source code TORCH_GEOMETRIC.NN.CONV.GRAPH_CONV .