Approved research

Detecting gene-gene interaction effect on blood cell traits by neural networks

Lay summary

The main purpose of Genome-Wide Association Studies is to understand the relationship between human genes and diseases. One approach is to analyze the relationship between human genes and metabolites or cell traits, which are highly related to most diseases. The effect of genes on traits can be divided into two parts: the main part (effect of each gene alone), and the interaction part (interaction effect between different genes). When the number of relevant genes is large (very common for most traits), the number of possible interactions grows rapidly, and detecting all possible interaction pairs can be intractable for traditional methods. This study focuses on developing modern neural networks algorithms which can detect gene-gene interactions on blood cell traits efficiently. Neural network algorithms have been successfully used in many high-dimension datasets such as images, speeches, and languages, and detecting statistical interactions by neural networks have also been widely studied in recent years. But in statistical genetics, few approaches have taken the advantage of 'Big Data' by using deep neural networks. This project will take about 1 year to develop related algorithms and test on blood cell traits data. Our work can also be generalized to study interactions on metabolites data, leading to interpretable relations between genes and diseases. The tool developed in this work has the potential to analyze other kinds of interactions, such as gene-environment interactions and gene-drug interactions, which can be a foundation for developing new drugs for many diseases in a genetic level.