Bayesian Semi-parametric Support Vector Machines Based on Multi-view Networks

Jin Ming¹Suprateek Kundu^2*

• ¹Emory University, Atlanta, United States
• ²University of Texas MD Anderson Cancer Center, Houston, United States

Brain networks have gained increasing recognition as potential biomarkers in mental health studies, but there are limited approaches that can leverage high-dimensional functional connectivity (FC) features for accurate classification. Our goal is to develop a novel Bayesian Support Vector Machine (SVM) approach based on high-dimensional networks as covariates that overcomes limitations of existing penalized methods, and to illustrate the advantages of this approach in classification analysis based on single connectomes and combining resting state and task-evoked connectomes. We develop a novel Dirichlet process (DP) mixture of Laplace priors on the coefficients in the Bayesian SVM model that is able to pool information across edges to determine differential edge-specific sparsity levels in an unsupervised manner. We develop different versions of the model that incorporates static connectivity from single as well as multiple fMRI experiments (multi-view analysis) and generalize the approach to incorporate dynamic FC features. The approach is implemented via a scalable Markov chain Monte Carlo (MCMC) algorithm that relies on a fully Gibbs sampler. We perform classification of intelligence levels using fMRI data from the Human Connectome Project (HCP), and a second Attention Deficiency Hyperactivity Disorder (ADHD) dataset. Our results clearly reveal the considerably greater classification accuracy under the proposed approach over state-of-the-art methods. The multi-view analysis integrating rest and task fMRI results in the highest accuracy. We provide concrete evidence that the novel Bayesian SVM provides an unsupervised and automated approach for network-based classification, which results in considerable improvements over penalized methods and parametric Bayesian approaches. Our work is one of the first to conclusively demonstrate the advantages of Bayesian SVM in high-dimensional network-based classification, and the importance of integrative multi-view network analysis.