A deep learning and experiment integrated platform for stable mRNA vaccines development

在所有的方法中，基于信使RNA(Mrna)的疫苗已经成为一种快速和通用的疫苗。 快速应对病毒大流行的候选方案，包括2019年冠状病毒病(新冠肺炎)。但m RNA 疫苗面临着关键的潜在限制。研究人员观察到，RNA分子往往自发地 降解，这是一个严重的限制--mRNA主干的一次切割就可以使mRNA疫苗失效。 目前，关于给定RNA的主干中哪里最容易降解的细节知之甚少 而设计超稳定的信使RNA分子是一个紧迫的挑战。在没有这种知识的情况下，mRNA 针对新冠肺炎的疫苗在制备、储存和运输方面将需要严格的条件。一个有希望的人 潜在的解决方案是深度学习，这是一种通用的数据驱动建模方法，已被证明占主导地位 在许多领域，包括计算机视觉、自然语言处理、蛋白质折叠和核酸特征 预测任务。在这项提议中，孙青博士的目标是将深度学习和实验相结合来预测mRNA 在室温下稳定的疫苗。通过采用包括自我注意在内的两种深度学习技巧 和卷积，她将创建可解释的端到端模型来预测新冠肺炎二次疫苗 结构，最后，她将使用一种快速合成的方法 产生信使核糖核酸疫苗来验证和进一步改进他们的深度学习模型。具体来说，这项研究 该方案的目标是：1)开发使用自我注意和卷积的深度学习模型，该模型 捕获长期依赖关系，预测RNA二级结构，并使用现有的 高精度和高效率的RNA二级结构数据集；2)对mRNA进行迁移学习 疫苗稳定性预测；以及3)验证和进一步改进模型性能 基于需求的信使核糖核酸生产系统。她将生产数百种mRNA疫苗序列并进行测试 他们在实验室中的稳定性作为数据集来验证和重新训练他们的模型。这个项目将作为一个 用于快速应对流行病的其他信使核糖核酸疫苗处理框架。二级结构 从这一提议中获得的预测知识也将有助于表征天然的和天然的合成的mRNA 科学和工程目的。

Among all approaches, messenger RNA (mRNA)-based vaccines have emerged as a rapid and versatile candidate to quickly respond to virus pandemics, including coronavirus disease 2019 (COVID-19). But mRNA vaccines face key potential limitations. Researchers have observed that RNA molecules tend to spontaneously degrade, which is a serious limitation - a single cut in the mRNA backbone can nullify the mRNA vaccine. Currently, little is known on the details of where in the backbone of a given RNA is most prone to degradation and design of super stable messenger RNA molecules is an urgent challenge. Without this knowledge, mRNA vaccines against COVID-19 will require stringent conditions for preparation, storage, and transport. A promising potential solution is deep learning, a general class of data-driven modeling approach, which has proved dominant in many fields including computer vision, natural language processing, protein folding, and nucleic acid feature prediction tasks. In this proposal, Dr. Qing Sun aims to combine deep learning and experiments to predict mRNA vaccines that are stable at room temperature. By adapting two deep learning techniques including self-attention and convolutions, she will create interpretable end to end models to predict COVID-19 vaccine secondary structures directly from sequence information and in the end, she will use a synthetic approach that rapidly generates mRNA vaccine to validate and further improve their deep learning model. Specifically, the research objectives of this proposal are: 1) to develop the deep learning model using self-attention and convolution, which capture long-range dependencies, to predict RNA secondary structures and to train the model using existing RNA secondary structure dataset with high accuracy and efficiency; 2) to employ transfer learning for mRNA vaccine stability predictions; and 3) to validate and further improve the model performance using experimental demand-based mRNA production system. She will produce hundreds of mRNA vaccines sequences and test their stabilities in the lab to serve as dataset to validate and retrain their model. This project will serve as a framework for other mRNA vaccine processing for rapid response to pandemics. The secondary structure prediction knowledge from this proposal will also help characterize natural mRNA and synthetic mRNA for natural science and engineering purposes.