Brain-like artificial networks are often referred to as a 'black box' because researchers do not know how they learn and make predictions. Researchers reported a way to peek inside the box and identify key features on which the computer relies, particularly when trying to identify complex DNA sequences.