基于深度学习的甲型流感病毒抗原性预测

甲型流感病毒可能导致季节性流感疫情甚至全球流感大爆发，这主要因为流感表面糖蛋白血凝素可以通过持续的抗原进化逃避宿主免疫，这也给疫苗决策提出了巨大的挑战。鲁棒的抗原相似性和抗原进化预测是应对疫苗挑战的理想方法。深度学习在很多领域显示出了它的优越性。因此，本项目以充分利用现有的遗传数据和抗原数据、从数据中发现规律为基本目标，以生物数据和深度学习相结合的策略研究抗原相似性和抗原进化的预测问题。预实验表明此研究具有可行性。项目整合现有分散的抗原数据并搭建数据共享平台，用网络模型对抗原相似性进行建模，用时间序列对抗原进化过程进行建模，进一步设计抗原相似性预测和抗原进化预测的深度神经网络模型，利用深度学习技术充分挖掘数据中隐含的关键信息，并利用这些信息实现抗原相似性和抗原进化的预测。项目发展抗原性预测的新模型、新方法和新平台，促进抗原数据的有效应用，为遗传变异和进化分析提供一种全新的思路和途径。

Influenza A virus can lead to seasonal and even pandemic outbreaks, mainly because the influenza surface glycoprotein hemagglutinin (HA) can evade host immunity through continued antigenic evolution. This has become a key challenge for vaccine decision making. Robust methods for antigenic similarity prediction and antigenic evolution prediction can meet this challenge. Deep learning has shown its superiority in many fields. This project aims to make full use of existing genetic data and antigenic data, discovers the patterns from the data, and studies antigenic similarity prediction and antigenic evolution prediction by combining biological data and deep learning. The pre-experiments show that this research is feasible. The project first integrates existing scattered antigenic data and builds a data sharing platform, models antigenic similarities with a network, models the antigenic evolution process with time series, further designs deep neural network models for antigenic similarity prediction and antigenic evolution prediction, learns the key information implicit in the data and predicts the antigenic similarities and the antigenic evolution by this information. The project develops new models, methods and a platform for antigenicity prediction, promotes the effective application of antigenic data, and provides a new way for genetic variation and evolutionary analysis.