An Integrated Computational-Experimental Platform for Holistic mRNA Sequence Design, Build, Test, and Learn

Anmol Seth, Adam Snider, Celeste B Marsan, Vanessa Vang

Keywords: mRNA design, codon optimization, active learning, contrastive learning, mRNA stability, translation efficiency, high-throughput screening, sequence optimization, design-build-test-learn, mRNA therapeutics

TL;DR: An integrated mRNA design-build-test-learn platform achieving 2.9-fold stability improvement, 61.8% expression gain, and 1.5-fold editing efficiency via novel algorithms, high-throughput wet-lab validation, and contrastive active learning.

Abstract: Messenger RNA therapeutics hold broad potential across infectious disease, oncology, and rare genetic disorders, yet designing sequences that simultaneously optimize stability, translation efficiency, manufacturability, and immunogenicity remains challenging due to the combinatorial size of sequence space and trade-offs between therapeutic objectives. Here we present an integrated design-build-test-learn platform that addresses these challenges through three contributions: (1) ChimeraFold, a codon-graph dynamic programming algorithm achieving 2.9$\times$ speedup and 522\% expanded sequence space coverage over prior methods; (2) a high-throughput automated wet-lab pipeline generating 29,000+ multimodal measurements; and (3) a contrastive learning framework for active learning-guided sequence selection. Evaluation on GFP and SpCas9 systems demonstrated 2.9-fold median improvement in stability, 61.8\% average enhancement in expression across four cell lines, and 1.5-fold improvement in gene editing efficiency over wild-type controls. The platform achieves mRNA half-lives up to 173 hours while preserving burst expression, and generalizes to commercial therapeutic targets in the hands of external partners with multiple-fold improvements in expression and stability.

Presenter: ~Anmol_Seth2

Format: Yes, the presenting author will attend in person if this work is accepted to the workshop.

Funding: No, the presenting author of this submission does not fall under ICLR’s funding aims, or has sufficient alternate funding.

Submission Number: 71