Understanding how plants interact with their colonizing microbiota to determine plant phenotypes is a fundamental question in modern plant science. Existing approaches for genome-wide association studies (GWAS) are based on the association analysis between host genes and the abundance of individual microbes, failing to characterize the genetic architecture of microbial interactions that are thought to a determinant of microbiota structure, organization, and function. Here, we implement a behavioral model to quantify various patterns of microbe-microbe interactions, i.e., mutualism, antagonism, aggression, and altruism, and map host genes that modulate microbial networks constituted by these interaction types. We reanalyze a root-microbiome data involving 179 accessions of Arabidopsis thaliana and find that the four networks differ structurally in the pattern of bacterial-fungal interactions and microbiome complexity. We identify several fungus and bacterial hubs that play a central role in mediating microbial community assembly surrounding A. thaliana root systems. We detect 1142 significant host genetic variants throughout the plant genome and then implement Bayesian networks (BN) to reconstruct epistatic networks involving all significant SNPs and find 91 hub QTLs. Gene annotation shows that a number of the hub genes detected are biologically relevant, playing roles in plant growth and development, resilience against pathogens, root development, and improving resistance against abiotic stress conditions. The new model allows us to better understand the underlying mechanisms that govern the relationships between plants and their entire microbiota and harness soil microbes for plant production.