Benchmarking Reverse Engineering Methods for Interaction Networks of Microbial Communities

In many ecosystems, from soil to ocean and lake water, including the human microbiome and saliva, there are hundreds of microbial species cohabiting the environment including bacteria, fungi, archaea and protozoa. They are connected in an intrinsic network of interactions defined by the shared resources and consumption dynamics. These types of interactions can be divided based on the type of influence one species has on another which can be positive, negative or neutral. By studying these interactions closely, an insight into the difference between healthy and diseased tissues can be gained which in turn could facilitate the development of suitable treatments for gut related diseases. With the advance of metagenomics sequencing techniques, the relative abundance of these species can be simultaneously experimentally recorded and studied. One of the steps in the analysis of the data obtained in this way is the attempt to construct the network of microbial interactions using a certain reverse engineering method. There are different methods available in the literature often based on correlation or mutual information. However, without knowing the original network their accuracy cannot be calculated. In order to be able to construct a ground truth network, the species abundance data used in this Thesis is simulated using the Community Simulator package. In addition, experimental noise is added to the data using the metaSPARSim simulator in order to obtain relative abundances that are as close as possible to the experimental data. This allows for a network comparison between the real and method-obtained interactions over multiple metrics as a comparison study. In addition, statistical results on the obtained data distribution are reported and compared with publicly available microbial data.