Deep learning unlocks sequence-divergent synthetic promoters to empower Streptomyces natural product engineering

Streptomyces are renowned for their unparalleled capacity to produce bioactive natural products, making them prime candidates for industrial antibiotic manufacturing. However, Streptomyces lags behind E. coli and yeast in genetic tool development, as its promoters often constrained by narrow strength ranges, poor availability, and minimal diversity, hindering advanced synthetic biology applications. Deep learning-powered sequence design revolutionizes genetic part engineering, offering unprecedented control over promoter performance. Leveraging a deep generative model, we computationally designed one billion promoter sequences, from which the 100 candidates were experimentally validated, with 92% exhibiting activity across an impressive 17,100% dynamic range (0.08 to 17.10-fold relative to the ermE∗p reference), including seven variants outperforming the strong kasO∗p benchmark. Notably, these synthetic promoters maintained robust functionality across four phylogenetically distinct Streptomyces hosts while displaying negligible sequence homology to native genome. Most significantly, their implementation dramatically enhanced titers of high-value compounds: a 28.6-fold increase for antifungal polycyclic tetramate macrolactams, 25.7-fold improvement for antibiotic daptomycin, and 6.1-fold boost for immunosuppressant rapamycin, demonstrating their transformative potential for metabolic engineering. This study established the first AI-generated promoter library for Streptomyces, underscore how machine learning can simultaneously achieve unprecedented design efficiency, extensive sequence diversification, and precision performance tuning.