In this project, we conduct data analysis experiments on Major Depression Disorder data provided by Le et al.'s research. This data is provided in two sets: gene expressions and demographic symptoms of subjects. Following their research, we recreated a similar scenario to understand the importance of different genes and symptoms. To measure the importance of these features, we trained different AI models and present the results of these experiments.
This project focuses on the analysis of Major Depression Disorder (MDD) dataset presented by Le et. al. in their research [1]. They adopt a co-expression network to detect the significant gene-expression modules on RNA-Seq data based on 79 healthy control (HC) and 78 MDD, in total 157 subjects. Within this co-expression network they utilize centrality metric to see the importance of individual gene expressions in different modules. Similar to this research, we conducted experiments on gene expression data, however; instead of using a co-expression network, we employed univariate, recursive and model-based feature selection methods to uncover the importance of different genes. Moustafavi et. al.'s work stated that there is no single gene expression association that indicates MDD, but joint gene expression yields significant results [2]. Parallel to these findings, there are overwhelming evidence that the MDD is heritable [3,4], therefore gene expressions can show importance on MDD diagnosis classification tasks. Following this idea, we conducted experiments with a subset of gene expressions from several feature selection method results. To achieve this goal, the feature selection plays a critical role since the joint gene expression is designed at this step of our experiment. There are several feature selection methods adopted in this project and their results are compared to see which gene expressions found significant by each method. The gene expressions that appear the most significant are grouped together. After the creation of the joint gene expression subset, we train different AI models to make binary classification on the subject's diagnosis, in other words, we experiment to see if the joint gene expression subset can help detect MDD. To explain our work in detail, first we present some background information.