Interpretable detection of novel human viruses from genome sequencing data

Jakub M Bartoszewicz; Anja Seidel; Bernhard Y Renard

Mapping predictions back to a target genome can be used both as a way of investigating a given model's performance and as a method of genome analysis. GWPA plots of wellannotated genomes highlight the sequences with erroneous and correct phenotype predictions at both genome and gene level, and nucleotide-resolution contribution maps help track those regions down to individual amino-acids. On the other hand, once a trusted model is developed, it can be used on newly emerging pathogens, as the SARS-CoV-2 virus briefly analyzed in this work. The methods presented here may also be applied to other biological problems outside of the field of pathogen genomics. However, experimental work and traditional sequence analysis are required to truly understand the biology behind host adaptation and distinguish true hits from false positives.