Abstract:
"In a little more than a century, viruses have emerged from being invisible to human knowledge to being recognized as ubiquitous biological forces shaping microbial ecosystems worldwide. Next-generation sequencing has accelerated virus discovery at an unprecedented rate, revealing vast viral diversity in the microbial world. However, this pace of discovery has created a critical knowledge gap: who do those viruses infect? This dissertation addresses this fundamental challenge by demonstrating that viral genomes contain detectable signatures of microbial host adaptation. Those genomic "battle scars" can be computationally decoded to predict virus-host interactions. Through analysis of these evolutionary fingerprints, this work shows that viral dependence on host cellular machinery creates measurable genomic signals that reflect specific adaptation patterns to their hosts. Leveraging these insights, a machine learning model was developed that predicts virus-host interactions with 92% accuracy. This tool, released as an open-source Python package, enables the mapping of virus-host interaction networks directly from sequence data. Application of this approach reveals the network architectures underlying virus-host interactions, with computational predictions compared with experimental approaches. Taken together, this body of work demonstrates how genomic signatures can be leveraged to predict virus-host ecological relationships, offering a new method for mapping virus-host networks from sequence data."