Ancestral Sequence Reconstruction as an Emerging Strategy for AI-Integrated Enzyme Engineering

Enzyme catalysis is central to biomanufacturing, but practical use is often limited by operational stability, substrate scope, and limited evolvability toward non-native transformations. Ancestral sequence reconstruction (ASR) provides alternative protein scaffolds and evolutionary sequence space for enzyme engineering. Some reconstructed enzymes show enhanced stability, broader substrate acceptance, or mutational robustness, although these properties depend on the family and reconstruction. This review examines three developments in the transition of ASR from a retrospective tool for studying evolutionary history to an artificial intelligence-integrated strategy for enzyme engineering. Tree-level multiplexed ASR (mASR) reconstructs ancestors across alternative phylogenies retained by statistical comparison. The resulting ancestral libraries capture evolutionary uncertainty and expand the accessible sequence space, with potential for functional exploration. Machine learning (ML), including protein language models, can support candidate prioritization and property prediction, but most ASR–ML demonstrations remain family-specific proof-of-concept studies. Generative and alignment-free methods are beginning to move the field beyond reconstructing sequences toward designing ancestral-like scaffolds for functions never sampled by natural selection. Together, these advances support the use of ASR to discover stable protein scaffolds, accelerate biocatalyst development, and investigate the molecular mechanisms of enzyme evolution.