Accelerating CO2 Biocatalysis through Multimodal AI-Guided Enzyme Prioritization and Spatial Confinement

Abstract Direct biological conversion of CO2 into value-added chemicals is severely restricted by sluggish enzyme kinetics and inefficient intracellular carbon flux. Here, we address these catalytic bottlenecks by synergizing multimodal deep learning, mechanistic enzyme engineering, and synthetic organelle assembly. Using CASKP, a TransKP-guided candidate-prioritization framework that integrates sequence representations with a docking-derived scalar feature, we prioritized and experimentally identified a highly active β-carbonic anhydrase, PbCA, for improving intracellular CO2 supply. Structural analysis and molecular dynamics simulations guided the design of the R71A and V112A variants. The R71A–V112A variant displayed a specific activity of 26.1 U/mg and increased the intracellular 13C-labeled 3-phosphoglycerate ratio 4.9-fold. The simulations were consistent with a more hydrated active-site environment and a strengthened Tyr108-centered hydrogen-bond network. To overcome diffusion and thermodynamic limitations, these optimized biocatalysts were encapsulated into hybrid carboxysomes and coupled with an ATP-generating carboxykinase. 13C tracing confirmed incorporation of inorganic carbon into l-malate, with 95.2% of malate detected as 13C-labeled isotopologues. Subsequent promoter optimization yielded an l-malate titer of 22.7 g/L. This multiscale approach provides an integrated strategy for heterotrophic CO2-incorporating biocatalysis.