Evolution and Active Learning Identify Terminal and Distal Determinants for Improved TAL in P ‐Coumaric Acid Production

p-Coumaric acid (p-CA) is a phenylpropanoid-derived platform molecule widely used in food, cosmetic, and nutraceutical products, and microbial biosynthesis represents a sustainable alternative to plant extraction. Tyrosine ammonia-lyase (TAL) catalyzes the single-step, cofactor-independent conversion of l-tyrosine to p-CA; however, natural TALs often suffer from low efficiency and limited robustness. Here, we established an "evolution-to-design" workflow integrating evolutionary analysis, terminal-region engineering, and active learning-guided semi-rational design to optimize TAL from Rhodotorula glutinis (RglTAL). Selection pressure analysis revealed strong positive selection at both termini. N-terminal truncation (28 residues) increased whole-cell activity by 31.0% and raised Tm by 1.0°C, while N-terminal replacements further enhanced activity, indicating an N-terminal conformational constraint. AI-assisted screening identified three distal substitutions (G30N, Q387A, L609A); the triple variant improved specific activity by 63.7% (0.31 U/mg) and raised Tm by 3.4°C. Combining these mutations with N-terminal truncation further increased activity to 0.36 U/mg. Molecular dynamics simulations suggest these gains arise from synergistic stabilization, active-site pocket tuning, and truncation-like N-terminal remodeling by G30N. This work provides an improved TAL biocatalyst and a generalizable framework bridging evolutionary insights and data-driven design for bioprocess-oriented enzyme optimization.