Statistical models and computational tools for gene-gene interaction analyses by utilizing multi-scale omics

点击翻译按钮获取中文摘要

Understanding the genetic basis of phenotypic changes and predicting phenotypes based on genotypes are long-standing goals of the field of genetics. Today, scalable tools, i.e., software that allows rapid analysis of very large data using limited memory in a personal computer, is an emerging requirement in the era of genomic big-data. The long-term goal of my research program is to develop novel statistical models and their scalable implementations to facilitate genotype-phenotype mappings and predictions. Background. Recent advances in high-throughput sequencing technologies including whole-exome sequencing, RNA-Seq (for transcriptome) and Bisulfite-Seq (for methylome) have created an excitement in genetics-related areas. However, there is a lack of tools that allow the seamless integration of multi-scale omics datasets for the precise prediction of phenotypes in a biologically relevant and meaningful context. In particular, gene-gene interactions have not been fully characterized and utilized in predictors. Moreover, in the coming big-data era, there is a lack of scalable tools that permit the effective statistical analysis of large datasets without requiring a machine with very large memory. Objectives. Building upon my previous work of identifying gene-gene interactions and implementing scalable software, I will focus on three short-term objectives: (1) Identify gene interactions using genotype-phenotype data by integrating multi-scale omics data; Bayesian Network and Frequent Itemset Mining will be integrated to achieve this goal. (2) Build a polygenic phenotype predictor that integrates gene interactions; an extension of Group LASSO will be implemented for this. (3) Create scalable software implementing the aforementioned statistical models using disk-based solutions, i.e., memory virtualization and huge-page techniques in computer science. It will store large data on disk while allowing rapid calculation as if the data resided in main memory. The multi-scale omics data from the 1,001 Arabidopsis Genomes Project and the multi-scale omics plant data generated in Alberta will be used. Impact. The proposed research will not only provide a new theoretical framework of statistical genetics, but will also furnish novel computational tools to assist experimental scientists carrying out gene mapping projects. Moreover, it will enable practitioners in agriculture and health to improve predictions of phenotype, benefiting Canadian food productions and the health system. Practically, my software represents a scalable solution for big-data analyses with minimum memory usage. This will be particularly relevant to many Canadian research groups that do not have immediate access to high-performance computing facilities. This program will train HQP to carry out bioinformatics and biostatistics analyses to fully utilize the future genomic big-data.