Identifying Influenza Virus Infection Severity and Outcome Signatures Through Artificial Intelligence-driven Analyses

摘要 流感及相关疾病仍然是经济和公共卫生负担的重要来源。每年， 约有300万至500万人感染严重流感病例，另有30万至50万人 全球范围内有更多的人死亡。在那些受到不成比例影响的人中，有一些人患有慢性病 糖尿病、肥胖症、心血管疾病、儿童和65岁及以上的个人。 例如，90%的季节性流感死亡病例和70%的流感相关住院病例属于 这个年龄段。还有一个重要的关联是，在感染后一周内发生心脏病发作 感染甲型或乙型流感病毒。为什么有些人更容易受到影响，可能是因为 免疫衰老、原有免疫、遗传、饮食、环境和其他潜在因素 疾病，这些疾病可能导致流感的严重程度。然而，科学家和临床医生仍然很难 为任何给定的参数或类别确定疾病严重程度的轨迹。这在很大程度上是由于分配 使用单一分析或方法得出的具有特定相关性的结果。考虑到可能存在的多种因素 影响疾病进展或免疫，一种将不同的数据集整合起来预测 结果将非常有益于指导临床实践。为了实现这一目标，我们假设 分析不同指标的多模式网络方法将从复杂数据集中识别符合以下条件的要素 预测流感疾病的结果。这些特征包括宿主、共生微生物和病毒因素， 以及它们之间与疾病和免疫相关的可识别的相互作用。这可能会揭示出新的 对流感病毒-宿主相互作用的洞察和改变临床实践。我们将利用已建立的 免疫分析、测序方法和元数据，以便利用现有方法 并将其应用于我们的新型生物信息学和人工智能工作流程。要做到这一点，我们将 一)生成一个全面的系统级多模式数据集，包括病毒因素和宿主因素，以评估 人类队列中不同的流感病毒感染严重程度特征，以及ii)利用多模式网络 分析和机器学习以识别预测疾病严重程度轨迹的特征和交互作用 由于流感病毒感染。

ABSTRACT Influenza and associated diseases remain significant sources of economic and public health burden. Every year, around three to five million people come down with severe cases of influenza with another 300,000 to 500,000 more who die worldwide. Amongst those who are disproportionately affected are individuals with chronic conditions such as diabetes, obesity, cardiovascular disease, children and individuals aged 65 years and older. For example, 90% of deaths from seasonal influenza and 70% of influenza-associated hospitalization belong to this age group. There is also a significant association between developing a heart attack within a week of getting infected with influenza A or B viruses. Why some people are more susceptible can be due to a multitude of factors such as immunosenescence, pre-existing immunity, genetics, diet, environment and other underlying diseases, which may contribute to the severity of influenza. Yet, it remains difficult for scientists and clinicians to determine the trajectory of disease severity for any given parameter or category. This is largely due to assigning an outcome with a particular correlate using a single assay or method. Given the multitude of factors that may influence disease progression or immunity, a method by which different datasets are integrated to predict the outcome would be highly beneficial in guiding clinical practices. Towards this goal, we hypothesize that a multimodal network approach in analyzing different metrics will identify features from complex datasets that are predictive of influenza disease outcome. These features include host, commensal microbial and viral factors, and identifiable interactions between them associated with disease and immunity. This will potentially reveal new insights into influenza virus-host interactions and transform clinical practices. We will utilize established immunological assays, sequencing approaches and metadata in order to take advantage of existing methods and infrastructures and apply it to our novel bioinformatics and artificial intelligence workflows. To do this, we will i) generate a comprehensive systems level multimodal dataset including both viral and host factors to assess differential influenza virus infection severity signatures in a human cohort, and ii) utilize multimodal network analysis and machine learning to identify features and interactions predictive of the trajectory of disease severity due to influenza virus infection.