QuBBD: Collaborative Research: Personalized Predictive Neuromarkers for Stress-Related Health Risks

冠心病（CHD）仍然是美国和其他后工业化国家成年人过早死亡的主要原因。预测冠心病的风险——尤其是对特定的个体——仍然是一个根本性的挑战。本项目将研究统计机器学习方法在多模态脑成像、行为、生物学和相关数据中的应用，以增强对冠心病风险的预测。它专门解决了在心理压力下人类大脑活动的特定模式是否可靠地预测已知的冠心病风险标志物的问题；也就是说，压力导致血压和动脉形态升高。从基础科学的角度来看，这项研究将促进我们对大脑与身体健康之间关系的机制理解。从公共卫生的角度来看，这项研究将有助于确定大脑活动的标记物，这些标记物可以客观地识别，并可能针对未来有冠心病风险的健康人群进行修改。将神经成像数据映射到冠心病风险的关键挑战在于能够非常精确地将观察到的心理应激反应回归到以数千体素记录的大脑活动时间序列上，并确定哪些大脑区域与回归最相关。传统的分析方法涉及通过使用特定的参数模型（如广义线性模型和固定的血流动力学响应模型）来形成粗略的时间摘要，这导致准确性较差。该奖项支持启动一个合作研究项目，该项目汇集了一个由统计机器学习、神经影像学和健康心理学研究人员组成的高度跨学科的团队，以解决以下两个目标：1)确定一个可推广的模型和神经标志物，以预测心理压力下基于神经动力学的心血管风险因素的个体差异。这将通过函数到实际和函数到函数套索回归的新方法实现；2)描述神经模式如何与其他生理和人体测量因素相结合，以个性化个体风险评分和神经生物标志物。该奖项由美国国立卫生研究院大数据到知识（BD2K）计划与美国国家科学基金会数学科学部合作支持。

Coronary heart disease (CHD) remains the leading cause of premature death among adults in the U.S. and other postindustrial nations. Predicting CHD risk - especially for a particular individual - remains a fundamental challenge. This project will investigate the application of statistical machine learning approaches to multimodal brain imaging, behavioral, biological, and related data to enhance the prediction of CHD risk. It specifically addresses the question of whether particular patterns of human brain activity during psychological stress reliably predict known risk markers of CHD; namely, stress-related rises in blood pressure and arterial morphology. From a basic science perspective, this research will advance our mechanistic understanding of how the brain relates to our physical health. From a public health perspective, this research will help to identify markers of brain activity that could be objectively identified and possibly targeted for modification in otherwise healthy people at risk for future CHD. The key challenge with mapping neuroimaging data to CHD risk lies in being able to very precisely regress observed psychological stress reactions on the time-series of brain activity recorded in thousands of voxels, and identify which brain regions are most relevant for the regression. Conventional analytic approaches involve forming coarse temporal summaries by committing to specific parametric models, such as a generalized linear model and a fixed model for hemodynamic response, that result in poor accuracy. This award supports initiation of a collaborative research project that brings together a highly cross-disciplinary team of statistical machine learning, neuroimaging and health psychology researchers to tackle the following two goals: 1) identify a generalizable model and neuromarkers that predict individual differences in cardiovascular risk factors based on neural dynamics under psychological stress. This will be enabled through novel methods for functions-to-real and functions-to-function lasso regression; 2) characterize how neural patterns can be integrated with other physiological and anthropometric factors to personalize individual risk scores and neuro-biomarkers. This award is supported by the National Institutes of Health Big Data to Knowledge (BD2K) Initiative in partnership with the National Science Foundation Division of Mathematical Sciences.