Hi, I am working with drug discovery in bachelor degree.
Now, I am using esm-2 to encode each node feature of protein and I use the graphormer to encode global the feature of ligand.
I hope, I can concatenate both feature matrix with add the padding 0 on missing feature categories in protein and ligand graph eg. protein feature (secondary structure) and ligand feature (node type) aren’t found in both of this graph.
If I add the 0 to fullfil this missing categories before I input in linear layer to compute dense feature before feed into other model will it cause model to confuse?
I have struggle in here:
- will model interpret 0 as the way that this label is the value that should be one of set of categories like (0 = beta sheet like categories from esm-2 then if I padding in the categories will it make model missing understand of ligand?)
- Do this method is effectively distinguish the different of protein or ligand feature.