In protein engineering, simultaneously improving multiple fitness attributes is a critical yet challenging goal, largely due to the vastness of sequence space, the multifaceted interplay among different traits, and the complexity of non-linear mutational effects (epistasis). To address this, we developed a data-driven evolutionary strategy that couples in silico deep learning with a wet-lab multi-objective selection workflow. By employing independent model fine-tuning for distinct traits, our approach facilitates navigating the fitness landscape to identify beneficial mutation combinations. We applied this strategy to T7 RNA polymerase (T7 RNAP), performing dual-fitness evolution to simultaneously enhance thermostability and activity at elevated temperatures. After five rounds of iterative evolution, we obtained T7 RNAP mutants exhibiting a melting temperature (Tm) increase of >10°C, a 60-fold enhancement in high-temperature activity, and a 70% reduction in by-product content. Validation in cell transfection demonstrated their potential for producing high-quality mRNA for industrial applications.
Jiang et al. (2026) studied this question.