Core Viewpoint - The rapid development of genomics presents unprecedented opportunities and challenges for characterizing protein functions, highlighting the limitations of traditional biochemical methods and the need for high-precision, quantitative approaches [3][4]. Group 1: Research Development - A machine learning model named AlphaCD was developed to accurately characterize 21,335 cytidine deaminases (CD), demonstrating high precision in predicting catalytic efficiency, off-target activity, target site window, and catalytic motifs [4][6]. - The research team experimentally characterized 1,100 APOBEC-like family cytidine deaminases fused with nCas9 in HEK293T cells, generating the largest dataset of experimental functional data for a single protein family to date [6][8]. Group 2: Model Performance - AlphaCD achieved high precision in predicting catalytic efficiency (0.92), off-target activity (0.84), target site window (0.73), and catalytic motifs (0.78) [6]. - The model was validated by subsampling 28 cytidine deaminases from the Uniprot database, with prediction accuracies of 0.84, 0.87, 0.75, and 0.73 for the respective features [6][8]. Group 3: Practical Application - The research team optimized the off-target site of one cytidine deaminase (A0A2R2Z4E4) using alanine scanning mutagenesis, creating a variant (A0A2R2Z4E4 E100A) that serves as a highly accurate and efficient cytidine base editor [8]. - This case exemplifies the application value of AlphaCD in high-precision, high-throughput protein function characterization and provides a paradigm for accelerating functional analysis of other proteins [8].