Fig. 5. Machine-learning-integrated multidomain combinatorial mutagenesis screen for resource-efficient protein engineering.

An overview of our study procedures and outcomes. We started with structure-guided design to select sites and residues for mutagenesis and built multidomain combinatorial variant libraries. We then run MLDE and tested embedding and model parameters to generate in silico predictions. We cross-validated experimental and MLDE predictions, which established parameters for accurate prediction of Cas9’s activity, fidelity, and targeting scope. The machine-learning-coupled multidomain combinatorial mutagenesis screening approach facilitates the identification of top-performing variants with much reduced experimental screening burden, increased hit enrichment, and enhanced resource efficiency.