Vector in the machine: how accurately can mosquito transmission of viruses be predicted by machine learning?

通过气候变化和人类旅行，许多蚊媒病毒正在迅速出现，并对全球健康产生重大影响。寨卡病毒和西尼罗河病毒在过去二三十年里出现了，乌苏图病毒也出现在整个欧洲，包括2020年入侵英国。对于这些病毒中的每一种，以前幼稚的蚊子物种和种群是新地区传播的驱动力。通过计算和入侵前预测一种新蚊子是否有能力传播病毒的能力，将极大地支持对新爆发的风险、准备和缓解的估计。我们的团队已经开发了机器学习框架来预测病毒/宿主的关联，广泛地适用于所有哺乳动物，特别是冠状病毒。正在进行的工作是使这些管道适应蚊媒病毒，这个项目将从实验上验证和改进这些机器学习管道。这个跨学科项目将第一个通过实验验证机器学习产生的蚊子/病毒能力预测，也是第一个基于实验验证的改进。使用机器学习对合格的蚊子/病毒组合的预测，并在CL3实验室通过蚊子活病毒感染来实验验证管道。蚊子群体将被喂食一种添加了病毒的血粉，孵化，并提取它们的唾液。唾液中存在病毒(空斑试验或qRT-PCR)表明有能力。使用这些结果，将识别机器学习管道中的优势/劣势，即它未能预测到什么？识别模式，并相应地改进管道。

Through climate change and human travel, many mosquito-borne viruses are rapidly emerging and having substantial impacts on global health. Zika and West Nile have emerged over the last 2-3 decades, and Usutu virus has emerged across Europe, including incursion into the UK in 2020. For each of these viruses, previously naive mosquito species and populations were the drivers of transmission in new areas.The ability to predict whether a novel mosquito is capable of virus transmission computationally and before incursion will greatly bolster estimation of risk, preparedness, and mitigation of new outbreaks. Our group has developed machine learning frameworks to predict virus/host associations, broadly across all mammals, as well as specifically for coronaviruses. Ongoing work is adapting those pipelines to mosquito-borne viruses, and this project will experimentally validate and refine those machine learning pipelines.This interdisciplinary project will be the first to experimentally validate machine learning-produced mosquito/virus competence predictions, and the first to make refinements based on experimental validation.Using machine learning predictions of competent mosquito/virus combinations, and experimentally validate the pipeline by live virus infection of mosquitoes in the CL3 lab. Colony mosquitoes will be fed on a virus-spiked blood-meals, incubated, and have their saliva extracted. Presence of virus in the saliva (plaque assay or qRT-PCR) demonstrates competence. Using these results, strengths/weaknesses in the machine learning pipeline will be identified i.e. what was it failing to predict? Identifying patterns and refine the pipeline accordingly.