Engineering of CRISPR-Cas PAM recognition using deep learning of vast evolutionary data

Area of Science:

• Biotechnology
• Genomics
• Molecular Biology

Background:

• CRISPR-Cas enzymes require protospacer-adjacent motif (PAM) recognition for genomic targeting.
• PAM specificity limits the range of sequences editable by CRISPR-Cas systems.
• Protein engineering offers a route to tailor Cas enzymes for novel PAM recognition.

Purpose of the Study:

• To develop an AI-driven tool, Protein2PAM, for predicting Cas protein PAM specificity.
• To utilize Protein2PAM for engineering Cas enzymes with altered PAM recognition capabilities.
• To demonstrate the application of machine learning in customizing CRISPR-Cas systems for genome editing.

Main Methods:

• Developed Protein2PAM, an evolution-informed deep learning model trained on over 45,000 CRISPR-Cas PAM sequences.
• Applied in silico deep mutational scanning to identify key residues for PAM recognition in Cas9.
• Computationally evolved Nme1Cas9 variants using Protein2PAM to broaden PAM recognition.

Main Results:

• Protein2PAM accurately predicts PAM specificity across Type I, II, and V CRISPR-Cas systems.
• Identified critical residues for PAM recognition in Cas9 without relying on structural data.
• Generated Nme1Cas9 variants with expanded PAM recognition and up to 50-fold increased cleavage rates in vitro.

Conclusions:

• Protein2PAM is the first successful application of machine learning for customizing Cas enzyme PAM recognition.
• This approach enables the engineering of Cas enzymes for specific, personalized genome editing applications.
• The findings pave the way for broader and more precise genomic targeting using CRISPR-Cas technology.