Overview

To ensure the final RNA therapeutics achieve high efficacy, specificity, safety, and predictability, and to accelerate wet-lab development, dry-lab approaches have been employed to construct multi-scale, integrated models from both molecular design and system prediction perspectives. These form a coherent closed loop and feedback cycle with wet-lab experiments.

To address limitations in existing 5′UTR design tools—such as limited search space, delayed functional prediction, and difficulty in handling variable-length sequences—we independently developed a fully automated "Generate–Score–Validate" pipeline. This integrates the high-performance UTRGAN model (which overcomes length constraints and significantly enhances expression) with the UTR-Insight model (enabling accurate scoring and screening). We also developed a workflow and web platform that allows users to perform closed-loop operations simply by specifying conditions, greatly lowering the barrier to use. Furthermore, we introduced a novel reverse optimization model to tackle the disconnection between sequence generation and functional performance, directly outputting high-quality sequences. Collectively, these efforts transform complex technologies into efficient, practical tools, providing superior 5′UTR sequences for our project's mRNA design.

To address the challenges of low efficiency and poor performance in the artificial design of toehold switches for eukaryotic genes, we drew inspiration from the TrigGate platform proposed by the iGEM 2022 TAU team. Through reproduction and code analysis, we developed an improved strategy that utilizes a genetic algorithm to expand the seed space. While maintaining efficiency and precision, this approach transforms the original model's fixed seed—limited to generating a finite set of sequences—into a system capable of generating an unlimited number of sequences, significantly enhancing sequence diversity.

To tackle the critical yet poorly characterized impact of the tumor hypoxic microenvironment on drug delivery and efficacy, we innovatively constructed a spatially resolved pharmacokinetic model. This model dynamically couples the heterogeneity of the hypoxic microenvironment in liver cancer with the delivery and expression processes of mRNA-LNP drugs. By quantifying two key mechanisms—hypoxia-enhanced vascular permeability and acid-catalyzed degradation—we achieved precise prediction of the spatiotemporal distribution and metabolic behavior of the drug across different hypoxic regions within the tumor. The model not only offers a computational basis for optimizing dosing strategies but also supports flexible multi-parameter adjustment via an interactive visualization platform, providing a powerful theoretical tool and design guidance for overcoming hypoxia-related barriers in mRNA therapy for liver cancer.

Together, these three components advance the end-to-end optimization of mRNA and gene therapies—from sequence design to in vivo expression—providing practical and robust tools to support project progression.