Learning from multi-level behaviours in agent-based simulations: a Systems Biology application