The paper describes the development of Protein2PAM, a deep learning framework designed to predict and engineer the protospacer-adjacent motif (PAM) specificity of CRISPR–Cas enzymes. By training on a massive dataset of over 45,000 sequences known as the CRISPR–Cas Atlas, the model identifies critical protein-DNA interactions without requiring complex structural data. Researchers used this tool to perform in silico mutagenesis, successfully designing Nme1Cas9 variants with customized or broadened recognition capabilities. These engineered enzymes achieved up to a 50-fold increase in cleavage rates compared to wild-type versions, significantly improving genomic targeting flexibility. Ultimately, this machine learning approach offers a rapid, scalable alternative to labor-intensive experimental methods for optimizing gene editing tools.

References:

• Nayfach S, Bhatnagar A, Novichkov A, et al. Customizing CRISPR–Cas PAM specificity with protein language models[J]. Nature Biotechnology, 2026: 1-10.