Establishing Multimodal Brain Biomarkers Using Data-driven Analytics for Treatment Selection in Depression

Project Abstract Major depression is the leading cause of ill health and disability worldwide according to the World Health Organization. Although significant progress has been made in understanding the disease and developing treatments, antidepressants, as the treatment mainstay, are effective for only about 50% of patients, in part due to the neurobiological and clinical heterogeneity in depression. Developing advanced data-driven techniques by leveraging machine learning with large-scale multimodal neuroimaging data from randomized clinical trials provides us a unique opportunity to explore brain biomarkers to identify treatment-predictive neurobiological phenotypes. Establishing such biomarkers is crucial for reducing the need for multiple drug trials and expediting remission by sharpening the search for treatment targets. However, integrative analysis of multimodal data for identifying biomarkers and differentiating individual responses to treatment in depression remains highly challenging and underexplored. In this proposal, we will develop new data-driven analytical tools to quantify multimodal moderators and signatures jointly from pre-treatment functional magnetic resonance imaging (fMRI) and electroencephalography (EEG) data for the prediction of treatment response to antidepressant medication. In Aim 1, we will identify multimodal moderators of treatment effect using data from the Establishing Moderators and Biosignatures of Antidepressant Response for Clinical Care (EMBARC) trial. A canonical correlation analysis-based data-driven model will be designed to extract combined features that fuse together complementary information from both fMRI and EEG modalities. Intent-to-treat prediction linear mixed models will be used to probe multimodal moderators of antidepressant sertraline versus placebo treatment response. In Aim 2, we will build a supervised latent space model that unifies the feature fusion and predictive modeling and apply it to quantify multimodal brain signatures that can predict individual treatment responses to sertraline versus placebo medication. In Aim 3, we will recruit 50 depressed patients as an independent cohort undergoing sertraline treatment to optimize and validate the identified multimodal biomarkers. Both fMRI and EEG will be collected at baseline followed by treatment with the antidepressant medication sertraline (in a manner paralleling EMBARC procedures) and clinical assessment of outcomes. We will release the developed software tools and collected data to be publicly available to the research community to facilitate multimodal neuroimaging studies in other mental disorders.