Dustin Scheinost - US grants

Project Summary (Abstract) The goal of this application is to develop, test, and disseminate multi-context (command line, desktop, server, and web applications) software for robust and reproducible neuroscience image analysis from data across mul- tiple scales (two photon microscopy, mesoscale Ca2+ optical imaging, small animal fMRI, and human fMRI) and species (mice, rats, and humans). In particular, we will provide tools for computation and visualization of con- nectomes (connectivity matrices) for such image data sets that will include both modality speci?c preprocessing (e.g. motion correction, nonlinear registration, noise removal) and species speci?c atlases and parcellations to create regions of interests for the computation of connectomes. Our goal is to provide software that can be used by end-users of different technical skill levels ranging from: (i) a limited technical background, whereby an investigator may simply put the data in a Dropbox folder and then use a web-browser to run the software from any computer (or high end tablet), to (ii) a more technically sophisticated background, whereby an investigator may use aspects of our software as command line scripts mixed with other custom tools to create a customized processing pipeline of their own design. In the ?rst scenario, we eliminate any need to download, con?gure, and install software. Only a modern web browser (e.g. Chrome, Safari, or Firefox) and a reasonably powerful computer (as all processing will be done locally) is required. A critical component of the proposed work is that the software will be designed from the ground up to enable robust and reproducible processing. As part of this design, this work will create output ?le formats that will store not only the results but also all necessary meta- data (software version, operating system, input data and parameters) to enable a different researcher to cleanly reproduce the results of someone else. We propose four speci?c aims: (1) Extend current algorithms to accept data from multiple modalities and species, (2) Design and implement multi-context desktop and web applica- tions, (3) Validate the translated algorithms and test the overall software, and (4) Document and distribute the software and train and engage end-users. The signi?cance of this proposal is that it will create software to ro- bustly and reproducibly analyze complex neuroscience imaging data across scales and species. The innovation lies both in the functionality proposed and in the multi-context design that will make the software accessible to end-users of different skill levels in different contexts (command line, server, desktop, and web applications) with robust cloud integration and output formats designed explicitly for reproducibility.

Project Summary (Abstract) A primary challenge facing functional neuroimaging is the translation of research findings to the clinical setting. In To predictive part, fMRI has struggled as a clinical tool due to the lack of functional phenotypes that characterize patients. address this, we have developed connectome-based predictive modeling (CPM) to identify and validate models of behavior/symptoms based on functional connectivity data. The promise of this approach is that by developing predictive models based on the functional organization of an individual's brain, we may be able to extract a rich connectivity phenotypes to aid in the clinical characterization of patients. This approach has the potential to improve our ability to categorize patients in otherwise heterogeneous groups and monitor the effectiveness of treatment interventions. To do this, modeling generalize methods across multiple behaviors, symptoms and diagnostic groups. are needed that are designed to In this proposal, we will push forward several major developments in CPM focused on generating transdiagnostic models for three specific behaviors (attention, working memory, and fluid intelligence) and factors from clinical tests, that will lead to functional phenotypes. We will collect a battery of continuous performance tasks in a spectrum of (N=300) individuals. We propose three specific aims: (1) To characterize node-boundary x dimensional construct effects; (2) To preform unidimensional and multi-dimensional CPM to predict RDoC constructs; (3) To evaluate the extent to which subjects with similar functional phenotypes cluster into symptom based or DSM-5 categorical clusters. This aim will also allow us to investigate the functional networks that vary with symptom and to investigate categorical subtleties in these symptom based phenotypes. The significance of transdiagnostic predictive models of behavior from functional connectivity data lay in their ability to delineate clinically relevant information from any individual (i.e. patient or control). The current lack of transdiagnostic predictive limits implications collection the clinical utility of f MRI in translating fMRI into a viable clinical tool. T of a novel trans-diagnostic data set to be made , providing a framework for, and generating, these models could models have important he innovation of this proposal is fourfold: 1) the publicly available ; 2) the development of an approach to generate personalized functional atlases to account for individual differences in anatomy; 3) the development of methods to delineate meaningful functional phenotypes to assess symptoms, and 4) to provide a means for comparing alignment of subjects on symptom dimensions versus DSM-5 categories using these functional phenotypes. These developments will be validated using a combination of novel data to be collected here as well as 3 publicly available data sets. The final deliverables will yield tools for measuring functional phenotypes reflecting symptom scores suitable for an individualized approach to medicine.

Abstract Alcohol initiation at an early age is associated with numerous negative outcomes, including a significant increase in the risk of developing an alcohol-use disorder later in life. Vulnerability for early misuse and other problematic alcohol use behaviors have been linked to individual differences in brain function. However, few studies have sought to identify brain-based predictors (?neuromarkers?) of alcohol use behaviors in youth. Identification of brain-based predictors of alcohol use behaviors in youth is essential for the development of more effective early prevention and intervention efforts. This proposal combines machine learning and longitudinal modeling approaches to 1) identify neural networks predictive of early alcohol initiation and misuse and 2) chart the developmental trajectories of these networks over time in a large sample of youth (N>3,000) using data from three unique, proprietary and completed datasets. Neural networks conferring vulnerability for alcohol use behaviors during adolescence will be identified using connectome-based predictive modeling (CPM). CPM is a machine-learning method of generating behavioral predictions from individual patterns of brain organization; i.e., functional connectivity matrices. Unlike traditional machine learning approaches, CPM is entirely data-driven and requires no a priori selection of brain regions or networks. As such, CPM is both a predictive tool and a method of identifying networks that underlie specific behaviors; i.e., neuromarkers. CPM has been successfully used to predict complex behaviors including future abstinence and other addiction-relevant phenotypes. This proposal will use CPM to identify neuromarkers of alcohol initiation and predict transitions to risky drinking in youth (AIM 1). Quantification of changes in brain function, e.g., growth curve trajectory analysis, is central to the characterization of developmental phenomena. Analyses of developmental trajectories can be used to identify particularly sensitive growth periods, detect variations that may signal risk, define modifiable targets, and monitor the impact of environment and interventions on development. While extant data indicate alcohol-related alterations in neural development, very few studies have assessed interactions between neurodevelopmental trajectories over time and alcohol-use behaviors. Developmental trajectories of identified networks in relation to alcohol use behaviors over time will be assessed using multilevel modeling (AIM 2). This proposal represents the first attempt to identify neural networks predictive of alcohol-initiation and risky drinking using a wholly data- driven, machine learning approach in a large sample of youth and does so using existing data. This is a critical step toward identifying a reliable predictor of alcohol initiation in youth and will shed light on individual difference factors representing vulnerability for misuse. Such predictors are needed to understand the developmental trajectories of alcohol phenotypes and to inform early risk models and preventative intervention efforts.

PROJECT SUMMARY The misuse of opioids, opioid addiction and overdose are a serious national public health crisis?the opioid epidemic?that despite increased scientific, clinical and government attention, continues to grow. Methadone is a generally effective treatment for opioid use disorder, however relapse rates remain high, and risk of overdose is greatest during relapse. There is a need for improved mechanistic understanding of the factors that contribute to opioid relapse to improve our understanding of opioid use disorder and its treatment. Using connectome-based methods (i.e., functional connectivity) in functional magnetic resonance imaging (fMRI), we recently identified a large-scale brain network that predicted opioid relapse from both resting and task states. Connectome-based methods enable data-driven characterization of whole brain networks related to behavior that might be better suited to describe complex clinical phenomena (e.g., opioid relapse). Building on prior work indicating the utility of real-time fMRI neurofeedback to test brain activation patterns related to specific functions and individual abilities to regulate these functions, the proposed project will use connectome-based neurofeedback to target patterns of functional connectivity within our recently identified ?opioid abstinence network?. This information is critical to improve understanding of mechanisms of opioid relapse. Individuals on methadone will be randomized to receive either active (n=12) or sham (n=12) connectome-based neurofeedback at 3 weekly scanning sessions including feedback and transfer runs. Additional baseline and follow-up scans will include resting state and reward and cognitive task runs. Craving, negative affect and opioid use will be measured weekly and at 1-mo follow-up. Based on our pilot data, connectome-based feedback will be targeted at the opioid abstinence network and we hypothesize that increased connectivity in this network will be associated with improved clinical outcomes. Aim 1 will test the hypothesis that active feedback is associated with reduced opioid use from baseline to follow-up scans (Aim 1a) and at 1-mo follow- up (Aim 1b). Aim 2 will test the hypothesis that active feedback is associated with increased opioid abstinence network connectivity in resting state (Aim 2a) and task (reward, cognitive) state (Aim 2b) versus sham feedback, as in our pilot work. Aim 3 will test the hypothesis that active feedback is associated with greater improvements in clinical features of opioid use disorder (craving, negative affect) than sham feedback (Aim 3a) and that increased opioid abstinence network connectivity will correlate with these improvements (Aim 3b). Overall, this project tests a potentially transformative hypothesis relating large-scale brain network dynamics to outcomes in opioid use disorder, and tests a highly innovative method for real-time fMRI neurofeedback from the opioid abstinence network to improve clinical features of opioid use disorder. This project will provide unprecedented insight into the functional neurobiology of opioid relapse and more generally has the potential to transform existing real-time fMRI paradigms in addictions.