Press Release
25 August 2026
A clearer view of what AI learns from DNA
A team from the Zeitlinger Lab developed PISA, a new method for visualizing what deep-learning models learn from DNA at single-base resolution. By separating experimental bias from biological signal, the team was able to train a more focused model that revealed previously hidden DNA sequence features linked to nucleosome positioning and the genome’s larger three-dimensional organization. The work also shows how model interpretation can guide new experiments and help biologists move from AI prediction toward biological mechanism.
Read Article
