Dynamically repairing and replacing neural networks: using hybrid computational and biological tools