Linking endotype and phenotype to understand COPD heterogeneity via deep learning and network science

摘要/摘要 慢性阻塞性肺疾病(COPD)是全球第四大死亡原因，导致 巨大的公共卫生负担。慢性阻塞性肺疾病的临床表现非常不同， 病程受多种内源性和外源性因素的影响。寻找具有以下特征的患者群 相似的病理生物学对于准确预测疾病进展和发展是至关重要的。 个性化治疗。目前，临床研究在区分患者的基础上存在分歧。 根据他们的表型特征，如肺功能，加重频率/强度，是否存在 肺气肿(临床亚型)，或根据其生物样本的分子组成进行评估 通过多组学分析(分子亚型)。尽管提供了关于不同分组的一些见解 对于COPD患者，在这两种分类方法之间几乎没有发现一致。因此， 病理生理过程、暴露和它们的表型后果之间的联系目前 不清楚。在这一应用中，我们建议使用深度神经网络体系结构来集成表型和 COPD受试者的基因组数据，并构建描述表型和 同时观察患者的分子特征。这些档案将被用来对患者进行分类，以找到关节 慢性阻塞性肺疾病的临床和分子亚型(内型)并预测5年的疾病预后 跨度。我们将提取每个内型的特征临床和分子特征，以获得内型- 并将其与慢性阻塞性肺疾病的临床表现联系起来。最后，我们将发展网络-- 了解与每种内型相关的关键分子途径和调控因子的方法。 要实现这个计划中提出的目标，将需要一套独特的技能，跨越生物学、网络 科学、机器学习和肺部疾病生物学。尽管马约里诺博士过去的职业生涯轨迹 使他为拟议的研究做好准备，促进我们目前对COPD异质性的理解是一种 具有挑战性的任务，需要在特定领域进行进一步培训。Maiorino博士已经开发出一种全面的 以肺部疾病生物学、组学数据集成和高维为重点的培训计划 统计数据。Maiorino博士将利用钱宁分部提供的丰富的智力环境 网络医学和哈佛医学院教授参加课程并定期与HIS 导师和顾问委员会成员。总之，Maiorino博士的培训和研究计划将使他能够 扩展他目前的技能，并发展成为一名独立的调查员，为促进 慢性阻塞性肺疾病的精准医学。

Summary/Abstract Chronic obstructive pulmonary disease (COPD) is the 4th leading cause of death worldwide, resulting in an immense public health burden. The clinical manifestations of COPD are extremely heterogeneous, and disease course is affected by numerous endogenous and exogenous factors. Finding groups of patients with similar pathobiology is crucial for the accurate prediction of disease progression and the development of personalized treatments. Currently, clinical research has been divided in the discrimination of patients based on either their phenotypic features, such as lung function, exacerbation frequency/intensity, presence of emphysema (clinical subtyping), or on the molecular compositions of their biological samples, as assessed through multi-omics assays (molecular subtyping). Despite providing some insights on different groupings of COPD patients, little agreement has been found between these two classification approaches. As such, the connection between pathophysiological processes, exposures, and their phenotypic consequences is currently unclear. In this application we propose to use deep neural network architectures to integrate phenotypic and genomic data of COPD subjects and construct integrated patient profiles that describe both the phenotypic and molecular features of the patient simultaneously. These profiles will be used to cluster patients to find joint clinical and molecular subtypes (endotypes) for COPD and to predict disease outcomes across a 5-year time span. We will extract the characteristic clinical and molecular features of each endotype to obtain endotype- specific biomarkers and connect them to clinical manifestations of COPD. Finally, we will develop network- based approaches to understand the key molecular pathways and regulators associated with each endotype. Achieving the objectives proposed in this plan will require a unique set of skills that span biology, network science, machine learning, and lung disease biology. Although Dr. Maiorino’s past career trajectory has prepared him well for the proposed research, advancing our current understanding of COPD heterogeneity is a challenging task that will require further training in specific areas. Dr. Maiorino has developed a comprehensive training program focusing on pulmonary disease biology, omics data integration, and high-dimensional statistics. Dr. Maiorino will take advantage of the rich intellectual environment offered by the Channing Division of Network Medicine and Harvard Medical School to attend courses and participate in regular meetings with his mentors and advisory board members. Altogether, Dr. Maiorino’s training and research plan will enable him to expand his current skillset and to develop into an independent investigator contributing to the advancement of precision medicine in COPD.