De novo design of insulated cis-regulatory elements based on deep learning-predicted fitness landscape.

Abstract

Precise control of gene activity within a host cell is crucial in bioengineering applications. Despite significant advancements in cis-regulatory sequence activity prediction and reverse engineering, the context-dependent effects of host cellular environment have long been neglected, leading to ongoing challenges in accurately modeling regulatory processes. Here, we introduce an insulated design strategy to purify and model host-independent transcriptional activity. By integrating heterologous paired cis- and trans-regulatory modules into an orthogonal host cell, we established a controllable transcriptional regulatory system. Using a deep learning-based algorithm combined with an experimental data purification process, we achieved the de novo design full-length transcriptional promoter sequences driven by a host-independent activity landscape. Notably, this landscape accurately captured the transcriptional activity of the insulated system, enabling the generation of cis-regulatory sequences with desirable sequence and functional diversity for two distinct trans-RNA polymerases. Importantly, their activities are precisely predictable in both bacterial (Escherichia coli) and mammalian (Chinese hamster ovary) cell lines. We anticipated that de novo design strategy can be expanded to other complex cis-regulatory elements by integrating the deep learning-based algorithm with the construction of paired cis- and trans-regulatory modules in orthogonal host systems.