A Deep Learning Framework for Discovering Temporal Changes in Host-Microbiome Interactions

About This Grant

The pandemic has underscored the importance of studying host-microbiome interactions, which play a critical role in human physiology and disease pathogenesis. The human microbiome is involved in diseases such as Helicobacter pylori with gastric cancer, Mycobacterium tuberculosis with tuberculosis, and human papillomaviruses with cervical cancer. Despite this, the role of microbiota has often been overlooked in large consortium studies due to limited access to clinical samples, technical challenges in quantification and lack of computational methods to integrate the multi-omics data. Dr. Chao Zhang, a computational biologist with over a decade of experience, specializes in developing computational algorithms, software, and data repositories. His work applies these tools to a wide range of biological and medical problems. In this project, Dr. Zhang proposes a framework using advanced deep learning methods to explore host-microbiome interactions. Over the next five years, this project aims to achieve the following: 1) Development of a novel microorganism identification pipeline: This deep learning-based pipeline will enable efficient and accurate identification of microbiota from routine sequencing data, replacing traditional alignment-based methods. This will greatly improve efficiency and reduce computational costs. Additionally, statistical methods will assess the abundance of detected microorganisms, minimizing false positives. The pipeline will be applicable to any existing sequencing data, allowing for retrospective analyses of large-scale studies not originally designed for microbiome research, as well as enabling contamination detection. 2) Creation of a deep learning-based method to analyze host-microbiome associations: This method will evaluate relationships between microbiome composition or metabolic functions and various host molecular features—such as transcriptomes, immune profiles, methylation levels, mutation profiles, and clinical data. The goal is to create a publicly accessible resource for researchers to better understand the molecular effects of microbiome dysbiosis across different tissues and diseases. 3) Introduction of a framework for longitudinal data analysis and multi-omics data integration: With the growing availability of time-series data, particularly in microbiome studies, this deep learning framework will address challenges in analyzing such data and integrating multiple types of omics data. This approach will offer researchers a powerful tool to investigate how host-microbiome interactions evolve over time. The potential applications extend to areas such as developmental biology, cancer research, drug development, and aging studies. This project’s innovative methods have the potential to significantly advance our understanding of microbiome dynamics and their broader impact on human health.