Original title: Optimizing RNA yield using deep neural networks coupled to massively parallel screening
Dinghai Zheng, Justin Hong, Jun Wang, Adrien Villain +3 · August 2026
Making large quantities of messenger RNA cheaply is a major hurdle for new vaccines and therapies. Scientists make RNA in test tubes using starting DNA, but they do not fully understand how specific DNA sequences next to the promoter region affect the final RNA yield. To solve this, researchers built an integrated deep learning framework. They tested a library of 10^5 randomized DNA sequences to measure their RNA yields in parallel. Next, they converted these sequences into numeric formats and trained a convolutional neural network on the data. The model learned how sequence patterns influence output, predicting yields with a 0.94 Pearson correlation against real experimental tests. By scoring and ranking proposed designs on a computer before building them, this tool lets scientists prioritize the easiest RNAs to manufacture. This framework cuts experimental screening costs and speeds up RNA engineering cycles for vaccine and protein therapies.
You are baking bread and want to know how much dough a new recipe will make. The recipe lists ingredients, but the final volume depends on small details you have not tested, like which flour or how warm the kitchen is. You bake a wide range of loaves, measure each one, and learn which choices reliably rise high. In this paper, the researchers apply the same logic to RNA production: they create many DNA sequences, measure how much RNA each one yields, and train a model to predict yield from the sequence before running the reaction.
mRNA vaccines and therapies depend on producing large quantities of RNA cheaply. Yield depends on DNA sequences near the start of the gene, but that relationship has been hard to predict. This model scores new designs before lab work, letting researchers prioritize the most manufacturable candidates and cutting screening costs and development time.
The abstract doesn't say whether the model's predictions have been validated in an actual production run, nor does it explain the biological mechanism linking promoter-adjacent sequence to yield. The library was explored within a defined structural context, so generalizing to other sequence contexts or to specific vaccines, protein replacement therapies, and cancer immunotherapy remains an outlook rather than a demonstrated result.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Optimizing RNA yield using deep neural networks coupled to massively parallel screening — Dinghai Zheng, Justin Hong, Jun Wang et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.