A Multivariate Mediation and Deep Learning Framework for Genome-Connectome -Substance Use Research

点击翻译按钮获取中文摘要

Substance use and addiction are complex biopsychosocial disorders influenced by both genetic and environmental factors. A key challenge in addiction genetics research is to understand how multiple genetic variants interactively influence addiction traits through impacting the central nervous system. To address this challenge, we propose a large-scale mediation analysis framework to identify addiction-related gene-brain circuitry pathways, using nicotine addiction as the targeted disorder, although the platform will be readily applicable for other addiction-related disorders and phenotypes. We will fully leverage the complex and interactive interdependent relationships between the imaging-genetics data and perform multivariate statistical inference with simultaneously increased statistical power and reduce false positive rates. The results will precisely identify multiple sets of genetic variants that interactively alter brain functional and structural circuitries, and then influence nicotine addiction. We will further supplement the mediation results with deep learning algorithms to study how genetic variants non-linearly and interactively coordinate to influence nicotine addiction and explain the phenotypic variance. Novel network topology based convolutional and pooling functions will be developed to achieve optimal prediction accuracy of addiction traits using genome-connectome pathways. All models and findings will be carefully validated through multiple independent large-sample data sets of imaging-genetics studies for nicotine addiction for ensuring the replicability and reliability of our findings derived from this framework. We plan to produce a freely available and user-friendly software incorporating the mediation analysis framework and deep learning algorithms enabling the complex whole genome - connectome analysis for addiction genetics research.