Guiding AI to Improve the Functionality of the Reporter Protein NanoLuc

NanoLuc, a small and exceptionally bright luciferase, is widely used as a reporter protein in imaging, biosensing, and real-time monitoring of cellular processes. Improving NanoLuc's functionality—such as increasing its luminescent output, thermodynamic stability, and catalytic efficiency—has the potential to significantly enhance signal-to-noise ratios in biological assays. In this work, we leverage BayesDesign, a probabilistic machine learning framework, to explore a vast sequence landscape of NanoLuc variants. BayesDesign enables rational exploration of mutation space using a Bayesian optimization strategy, balancing exploration and exploitation to identify candidates likely to yield functional improvements. To ensure biological plausibility, region-sensitive constraints are applied during sequence generation, preserving structurally critical domains while allowing targeted mutational flexibility elsewhere. From thousands of generated variants, a small set of candidates is prioritized for experimental testing based on predicted functional enhancements.This work demonstrates the potential of combining Bayesian optimization with physics-based constraints to create high-performing protein variants.