(174cb) AI-Guided Protein Engineering: Achieving Multi-Property Optimization for Enhanced Biocatalysis in Pharmaceutical Manufacturing

Low biocatalytic activity in industrial enzymes leads to significant financial losses in pharmaceutical manufacturing due to inefficiencies and suboptimal yields. Traditionally, engineering enzymes to optimize multiple properties simultaneously—such as activity, stability, and selectivity—has been challenging and time-consuming. In this work, we present how Pando Bioscience leverages AI-guided protein engineering to significantly enhance enzyme performance in pharmaceutical biocatalysis. Pando developed a multi-modal AI model integrating structure-based, sequence-based, variant prediction, evolutionary, and task-specific approaches, all fine-tuned with high-quality proprietary data. Using this model, we achieved approximately a 14-fold increase in enzymatic activity while maintaining high conversion rates and enantiomeric excess in less than a month. These rapid improvements, accomplished swiftly and with freedom to operate, showcase the efficiency of our AI-driven approach. Additionally, we illustrate how AI can guide metagenomic sourcing of novel enzymes for therapeutic applications. Specifically, we identified enzymes capable of converting active toxic small molecules into inactive, non-toxic forms, opening new avenues for detoxification processes.