基于深度学习分析H5禽流感病毒人类适应性突变组合及潜在危害

The cross-species transmission of H5Ny from avian to human poses a huge threat to public health. A large number of studies have found that many single-site mutations lead to the adaptation of the virus to humans. However, our previous studies have shown that some experimentally validated human-adaption-associated mutations (such as A138V for HA, D701N for PB2, etc.) are not common in viruses isolated from humans. This shows that in addition to the single-site mutations at these star sites, there are other mutations or mutation combinations that we do not know, which can adapt the avian H5Ny to humans. Based on this, the project intends to use deep learning and joint follow-up experiments to verify a batch of mutation combinations related to H5Ny's adaptation process from avian to human. Firstly, two types of sequences were trained a convolutional neural network (CNN) and the network parameters are tuned automatically. Secondly, the trained CNN model is used to train the soft decision tree (“distilling” the classification knowledge learned by the CNN into the soft decision tree) to obtain a visual classification criteria; finally we will select 1-2 mutation combinations and verify them experimentally. The implementation of this project will shed light on how H5Ny adapts to humans from avian at genomic level and will provide important clues for future influenza warnings.

H5Ny禽流感病毒自禽到人的跨种传播对公共卫生构成了巨大的威胁。大量的研究发现了许多单点突变导致病毒对人类的适应。然而，我们前期的研究显示目前一些经过实验验证的与人类适应性相关的突变（如HA的A138V、PB2的D701N等）在感染人的病毒中并不常见。这就说明了除了这些明星位点的单点突变以外，还存在其他我们未知的突变或突变组合，可以使得禽源H5Ny病毒适应人类。基于此，本项目拟使用深度学习并联合后续实验验证，发现一批与H5Ny从禽到人适应过程相关的突变组合。首先使用卷积神经网络对这两类序列进行学习并进行网络参数调优；其次，使用该模型训练软决策树（将深度神经网络学习到的分类知识“蒸馏”到软决策树中）从而得到一个可视化的分类标准；最终选取1-2个突变组合，对其进行实验验证。本项目的进行将会使得对H5Ny如何在基因组层面适应人类有更深刻的了解，并将会对今后流感预警提供重要的线索。

病毒的变异及其可能出现的跨种传播威胁着人类的健康，因此有必要开发预警模型来对病毒的变异及对其传播能力的影响进行评估。本课题整合了进化信息、统计信息以及蛋白质结构信息等多维度信息，构建了一套流感病毒的预警模型，鉴定潜在的人类适应性位点。我们以H5亚型禽流感为例，鉴定出了10个与人类适应性显著相关的突变位点，其中大部分都是之前没有被发现的位点。这为今后评估新型流感病毒对人类的适应性提供了理论基础。此外，我们对模型进行了参数调整，也应用于新型冠状病毒传播的预警，揭示了新冠疫情早期的传播异质性以及评估了新冠水貂突变体的社区传播程度。