AI-Guided Design and Predictive Modeling of Synthetic Escherichia coli Promoters through Comprehensive -10/-35 Box Engineering.

Promoters are essential in transcriptional regulation, with the -10 and -35 boxes playing a critical role in determining their strength. Modulating these regions can effectively fine-tune promoter strength. However, the lack of a clear quantitative relationship between sequence composition and transcriptional output impedes the rational design of promoters. To address this, we developed a synthetic promoter library by varying RNA polymerase binding energies at the -10 and -35 boxes. The library was partitioned into four sublibraries with expression strengths spanning an 80-fold range. Using fluorescence-activated cell sorting followed by sequencing, we identified 20,799 distinct promoters. Analysis of this library uncovered distinct sequence-activity patterns, including a small subset of -35 box sequences that consistently conferred high transcriptional output across diverse -10 partners. Based on this, we developed an artificial intelligence platform that integrates a convolutional neural network for strength prediction (Pearson's r = 0.84) with a balanced generative adversarial network incorporating a gradient penalty for de novo promoter design. By coupling these models, we achieved a precise design of promoters with user-defined strengths (r = 0.85), establishing a bidirectional framework that links -10/-35 boxes to transcriptional activity through deep learning. This study expands the sequence diversity of functional -10 and -35 boxes in E. coli, provides a predictive platform for rational promoter engineering, and deciphers combinatorial motif interactions governing transcriptional regulation.