When you generate derived models as read/write, you'll get code which materializes the root entity from the dto, but not related entities. This is because DTOs often don't contain enough information to materialize the related entities (e.g. no pk values) and are often projections of the related entities, not the related entities themselves. After all, the derived models are not said to be 1:1 copies of the entities. So if you want to do this, you have to write the code yourself.
Are you editing entities through DTOs in a graph form btw? We found only use cases where root entities were edited through DTOs but not in graph form. If you do that, why not just transport the entities as a graph?