Computation and deep-learning-driven advances in CRISPR genome editing

Genome editing with CRISPR-Cas systems is revolutionizing medicine, molecular biology and biotechnology. In this Review, we discuss the contributions of deep learning-based structure prediction algorithms, physics-based simulations, neural networks, graph neural networks and generative models, including diffusion and large language models, in engineering and optimizing CRISPR systems and in understanding their mechanistic basis. We highlight the challenges and limitations to the transformative effects of computational modeling and tools in the context of the development of programmable genome editors for biomedicine and biotechnology.