New AI model weighs when protein look-alikes can be trusted - Bioengineer.org