Multi-Omic Systems Biological Analysis of Host-Microbe Interactions

Systems biology offers the opportunity to understand the complex mechanisms of various biological phenomena. The wealth of data that is produced, at an increasing rate, provides the potential to meet this opportunity. Here we take an applied approach to integrate multiple omic level data sources in order to generate biologically relevant hypotheses. We apply a novel analysis pipeline to model both, in concert, the microbial and transcriptomic signature from COVID-19 positive patients. We show patients may suffer from an increased microbial burden, with an increased pathogen potential. Gene expression evidence further shows patients may exhibit a compromised barrier immunity, owing to the dysfunctional mechanism underlying cilia-associated mucosal clearance. We apply a similar pipeline together with genomic variant data in black cottonwood, \textit{Populus trichocarpa}. Here we aim to understand host molecular mechanism that may play a role in microbial community structure. We characterize the microbial diversity from population wide leaf and xylem samples of 433 genotypes. From this information we derive microbial phenotypes for a Genome Wide Association Study (GWAS). We find significant associations between microbial taxa and genes involved in plant signaling, phytohormone response, epigenetic regulation, and biotic stress response among others. Population structure derived from the similarity of microbial communities across genotypes has associations with distinct taxa and genes involved in antagonistic phytohormone pathways. As a further step towards integrating multiple omic data sets, we develop a deep learning framework. Our framework allows for the modeling of heterogenous, noisy, and high-dimensional data. We use a transformer mechanism together with an autoencoder to embed the omic data into a common topology. By analyzing the attention weights, we can interpret how the topology relates to the original features. We profile our framework on several real biological data. Multi-omic data analyses in systems biology allow us to improve our current understanding of many biological phenomena. Such as generating hypotheses that may help develop therapeutics for COVID-19 patients. Our analyses also provide a wealth of information on avenues to explore regarding host-selection mechanism that may influence microbial diversity. Deep learning frameworks may also provide additional lines of evidence for hypothesis generation.