Deep Generative Model-Driven Design of Microbial Synthetic Promoters

A synthetic promoter is an artificially designed DNA sequence based on naturally occurring promoter elements, enabling more precise control of gene expression than natural promoters. Design of synthetic promoters with tunable expression levels is key to precise genetic regulation in microbes, supporting metabolic engineering, natural product biosynthesis, and diverse biotechnological applications. Recent advances in deep learning have made it possible to generate functional synthetic promoters using deep generative models (DGMs). Such approaches dramatically accelerate the traditionally labor-intensive and time-consuming process of experimental promoter design, enabling the efficient discovery of synthetic promoters. In synthetic promoter generation, three major types of DGMs have been predominantly employed: variational autoencoders (VAEs), generative adversarial networks (GANs), and diffusion models. VAEs reconstruct promoters through latent feature learning, GANs create realistic promoter sequences via adversarial training, and diffusion models iteratively denoise random inputs to generate high-fidelity synthetic promoters. This review outlines deep learning-based strategies for synthetic promoter design, encompassing data acquisition, promoter generation, and validation of promoters generated by DGMs.