Decoding Enzyme Sequence-Activity Relationships via Generative AI for Rational Enzyme Design

Project Summary Rational enzyme design holds the promise to transform biological research and therapeutic development by enabling the creation of efficient biocatalysts. However, current methodologies in rational enzyme engineering remain limited; computationally designed enzymes often fall short of natural enzymes and require extensive experimental optimization. A fundamental barrier to progress is the incomplete understanding of enzyme sequence-activity relationships. In fact, even the most advanced data-driven and physics-based models cannot reliably and systematically predict mutation effects on enzyme activity. Such a prediction is critical for designing optimal enzyme sequences for efficient catalysis. This limitation underscores the need for novel approaches. Generative AI offers a compelling solution to overcome the barrier, analogous to how language models like ChatGPT interpret semantics in human language. Generative AI can learn from sequences evolved in nature to predict protein function. In particular, the PI has pioneered the application of generative AI to predict enzyme activity using sequence data. Our research has demonstrated that natural sequence information can reliably predict the impact of mutations on enzyme catalytic activity, especially in active sites, enabling the successful optimization of multiple enzymes in biochemical assays. Additionally, we found that loop sequence within TIM- barrel scaffold enzymes can predict novel activity when integrated into de novo designs. These insights represent a new paradigm for studying enzyme catalysis, highlighting how different enzyme regions are evolved for catalytic efficiency in nature. This research program aims to further advance our understanding of enzyme sequence-activity relationships through generative AI, with the long-term goal of achieving fully rational enzyme design. In Direction 1, we will evaluate the generalizability of these relationships across diverse enzymes, laying the groundwork for future sequence-based enzyme engineering. Directions 2 and 3 will overcome the current limitations of black-box AI models by incorporating physics-based modeling to elucidate the sequence-activity relationships identified by generative AI. Specifically, Direction 2 will investigate how active sites are preorganized for efficient catalysis. Direction 3 will explore the role of loops in TIM-barrel enzymes and develop strategies to enhance catalytic activity by engineering these loops. Overall, this proposal aims to introduce innovative methodologies to link enzyme sequences with their catalytic activities, deepen our understanding of enzyme catalysis, and pave the way for fully rational enzyme design.