AI-powered multi-scale modeling of microbiome-host interactions

Prediction of organismal characteristics based on genetic signatures of an individual organism is a fundamental problem in biology. Recent advancements in the field indicate that microbiome, a community of microorganisms found in all environments on earth from human bodies to soils, plays an essential role in determining the function of ecosystems. Microbiome

generates a variety of metabolites that serve as messengers for the microbiome to communicate with its surroundings. It remains unclear how these messengers communicate. This research intends to decipher the molecular language of microbiome-environment interactions, thereby resolving a major enigma in life sciences. Understanding the fundamental rules governing

microbiome-environment interactions will facilitate the solution of pressing problems in agriculture, environment, energy, and human health/well-being. By modulating the function of the microbiome in plants, for instance, agriculture could become more productive while requiring less pesticides and fertilizers. Utilizing the microbiome's photosynthetic energy is a promising

alternative clean energy solution. Maintaining gut microbiome stability may improve human immunity. This multidisciplinary research involving biology, chemistry, and computer science will create opportunities for underrepresented students in life sciences to pursue careers in computer science and bioinformatics, fields in which racial and ethnic minorities are

woefully underrepresented. Metabolite-protein interactions (MPIs) play a major mechanistic role in maintaining microbiome communities and mediating microbiome-host interactions. Exploration of the global landscape of MPIs and the host gene expressions regulated by the MPI would fill critical knowledge gaps in

causal environment-genotype-phenotype associations. However, our understanding of MPIs and their regulatory pathways is limited due to the transient and low-affinity nature of MPIs. Existing experimental techniques for determining MPIs and their functional roles are time-consuming, biased to certain molecules, or applicable on a relatively small scale. The rapid growth of multiple

omics data provides us with new opportunities to predict MPIs across entire microbiomes and host genomes. However, due to noisiness, biasness, incompleteness, and heterogeneity of omics data sets, their potential has not been fully appreciated for developing accurate, robust, and interpretable predictive models for MPIs and their functions. The project will develop an

innovative AI-powered multi-scale modeling framework to predict biological functions of microbiome metabolites by integrating diverse data from chemical genomics, structural genomics, and functional genomics. Specifically, this research will develop a novel end-to-end meta-learning method following a sequence-structure-function paradigm to predict genome-wide microbiome

metabolite-host protein interactions, and innovative domain adaption methods to translate cell line screens to host systemic responses to microbiome metabolites by disentangling intrinsic biological signals from both biological and technical confounders. An immediate outcome will be a bespoke platform to infer novel biological relationships from noisy, biased, and incomplete data, which

will have broad applications in addressing many fundamental biological problems. Another far-reaching benefit will be the discovery of novel genetic, molecular, and cellular mechanisms of biological processes. Completing this project will provide a powerful computational framework to

correlate molecular interactions to physiological processes, and establish causal environment microbiome-genotype-phenotype associations. The results of the project can be found at https://github.com/XieResearchGroup/.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.