Connectome-Based Prediction of Addiction Severity

7.摘要 美国目前正面临着与使用多种类型的 物质.鉴于这一公共卫生危机，迫切需要确定跨诊断， 物质使用障碍（SUD）的生物标志物。在临床神经科学领域，传统的，群体水平的 功能性磁共振成像（fMRI）方法一直在努力识别预测性生物标志物， 成瘾在样本中复制，可能是由于（a）过拟合数据和（B）检查 不同的SUD。为了解决这个问题，本申请的第一个目标是利用最新的进展 在基于连接组的预测建模（CPM）中，一种数据驱动的机器学习方法， 功能连接模式（“神经指纹”），预测跨诊断，维度 在个体水平上衡量SUD严重程度。此外，与单一模态研究相反，NIDA 概述了研究生物学、心理学和生物学因素之间复杂相互作用的必要性， 环境领域（优先重点#1），以了解SUD脆弱性的实际复杂性 （重点关注#3）。根据这些优先事项，本申请的第二个目标是整合 SUD相关的全脑指标纳入成瘾脆弱性的多层次模型，包括完善的 心理和环境风险因素。拟议的研究将利用收集到的档案数据 作为两项独立研究的一部分（N = 242）。临床评估、自我报告数据和静息状态功能磁共振成像 数据收集自社区成年人的不同样本，非法药物使用率升高，43%的人符合 终生SUD的标准和81%的报告非法药物使用史。我们假设CPM衍生的 神经网络将在训练样本中出现与SUD严重程度相关的神经网络， 在另一个独立样本中预测SUD严重程度（目标#1）。这有望产生一个工作神经 该模型可以进一步测试并应用于对其他SUD严重程度进行高度个性化的预测， 看不见的样本接下来，我们假设这些网络将为多方面的 成瘾脆弱性模型，超越了既定的心理和环境风险 （目标#2）。跨多个领域的知识集成使该应用程序能够很好地 解决SUD的现实世界复杂性，从而为取得重大进展提供了机会， 精确医疗方法的发展。此次F31应用将为 申请人在两个关键领域的培训：（a）实施机器学习所需的技术技能 分离具有预测能力的大脑网络的方法，以识别SUD严重程度的个体差异 以及（B）多方面模型的概念化和设计，该多方面模型将来自多个 分析领域来描述成瘾脆弱性。

7. Abstract The United States is currently facing an epidemic of fatalities associated with the use of multiple types of substances. Given this public health crisis, there is an urgent need to identify transdiagnostic, dimensional biomarkers of substance use disorders (SUDs). In the field of clinical neuroscience, traditional, group-level functional magnetic resonance imaging (fMRI) approaches have struggled to identify predictive biomarkers of addiction that replicate across samples, possibly due to methods that (a) overfit the data and (b) examine different SUDs in isolation. To address this, the first objective of this application is to leverage recent advances in connectome-based predictive modeling (CPM), a data-driven machine-learning approach, to identify patterns of functional connectivity (“neural fingerprints”) that are predictive of a transdiagnostic, dimensional measure of SUD severity at the individual level. Further, as opposed to single-modality research, NIDA has outlined the need to examine the complex interactions of factors across biological, psychological, and environmental domains (Priority Focus #1) in order to understand the real-word complexity of SUD vulnerability (Priority Focus #3). In line with these priorities, the second objective of this application is to integrate SUD-related whole-brain metrics into multilevel models of addiction vulnerability that include well-established psychological and environmental risk factors. The proposed research will make use of archival data collected as part of two independent studies (N = 242). Clinical assessments, self-report data, and resting-state fMRI data were collected from a diverse sample of community adults, elevated on illicit drug use, with 43% meeting criteria for a lifetime SUD and 81% reporting a history of illicit drug use. We hypothesize that a CPM-derived neural network will emerge in relation to SUD severity in the training sample that can be used to accurately predict SUD severity in another independent sample (Aim #1). This is expected to produce a working neural model that can be further tested and applied to make highly-individualized predictions of SUD severity in other, unseen samples. Next, we hypothesize that these networks will contribute unique variance to a multi-faceted model of addiction vulnerability, above and beyond well-established psychological and environmental risk factors (Aim #2). The integration of knowledge across multiple domains makes this application well-situated to address the real-world complexity of SUDs, and thus presents an opportunity for significant advances towards the development of precision medicine methods. This F31 application will provide opportunities for the applicant’s training in two critical areas: (a) the technical skills necessary to implement machine learning approaches to isolate brain networks with predictive power to identify individual differences in SUD severity and (b) the conceptualization and design of multi-faceted models that incorporate factors from multiple domains of analysis to characterize addiction vulnerability.