MAG-based Discovery and Clinical Application in Antimicrobial Resistance and Cancer Immunotherapy

Abstract

The human gut microbiome is a critical determinant of host health, yet its dysbiosis presents significant clinical challenges. In the context of infectious diseases, the misuse of antibiotics has accelerated the emergence of antimicrobial-resistant (AMR) bacteria, transforming the gut into a reservoir for resistance genes. While surveillance of environmental and hospital samples has successfully identified these reservoirs, the specific functional roles of AMR bacteria within individual gut ecosystems remain poorly understood. Current metagenomic studies often fail to capture the strain-level dynamics and functional niches that allow specific key- stone species to persist and restructure the commensal community following antibiotic treatment. Similarly, in the field of oncology, the gut microbiome has emerged as a key modulator of Immune Checkpoint Blockade (ICB) therapy. Despite its potential, ICB efficacy is highly variable, with response rates hovering around 30%. Attempts to predict patient response using microbial signatures have been hindered by significant inter-cohort heterogeneity. Predictive models trained on specific populations often fail to generalize to external datasets, yielding poor accuracy due to regional variations in taxonomic composition. A fundamental limitation is the reliance on taxonomy-based analyses. The same species can harbor distinct functional gene repertoires across different clades, meaning that species identity alone cannot capture functional heterogeneity. A common thread connecting both challenges is the need for high-resolution functional analysis that goes beyond taxonomic classification. Whether driving antibiotic resistance or modulating immune responses, the clinical impacts of bacteria are often defined by their functional capacity rather than their taxonomic identity. My doctoral thesis addresses these limitations through Metagenome-Assembled Genome (MAG)- based computational analysis. In the first study, I performed strain-resolved metagenomic assembly to discover and characterize Extensively Acquired Resistant Bacteria (EARB). Through novel Community Power analysis, I revealed how EARB utilize specific metabolic advantages to reshape the gut ecosystem and outcompete commensal species, demonstrating that functional impact—not just taxonomic identity—determines bacterial dominance in post-antibiotic environments. In the second study, I applied the same MAG-based pipeline to generate high-resolution functional profiles for cancer immunotherapy prediction. I developed imPath, a novel method that converts hierarchical functional data (EC numbers) into 2D image representations, preserving the 4-level enzymatic hierarchy for deep learning analysis. Critically, I demonstrated that MAG-derived functional profiles (AUC 0.72) outperform taxonomy-based approaches (AUC 0.69) in predicting ICB response. Integration of both modalities through Gated Multimodal Fusion achieved the highest performance (AUC 0.74), with functional information serving as the primary predictive axis and taxonomy providing complementary regional context. By establishing MAG-based functional analysis as a unifying methodology across infectious disease and oncology, this research demonstrates that the key to understanding complex host-microbiome interactions lies in analyzing what microbes do, not just who they are. This work provides comprehensive insights into microbial resilience and establishes a robust foundation for precision medicine.