Neural Network-Assisted Humanisation of COVID-19 Hamster Transcriptomic Data Reveals Matching Severity States in Human Disease

The recent coronavirus disease 2019 (COVID-19) pandemic has highlighted the need for appropriate models to evaluate therapeutic options applicable in a clinical setting. Though animal models are particularly valuable for studying host-pathogen interactions, approaches enabling comprehensive matching of experimental data to humans are still scarce. High throughput approaches such as single-cell transcriptomics (scRNAseq) have the ability to dissect molecular and cellular changes arising during inflammation, however computational methodologies robustly linking and evaluating interspecies changes in model organisms remains lacking. Here, we introduce a neural network using variational autoencoders capable of mapping temporal disease states of two hamster models – presenting moderate (Syrian hamster) or severe (Roborovski hamster) disease courses upon infection – to human COVID-19 severity ranks. Quantification of individual cell similarities confirmed that the transcriptional state of most Syrian hamster cell types best matched that of patients with moderate disease progression. Roborovski hamsters, which develop fatal outcomes upon SARS-CoV-2 infection, shared the highest similarities in neutrophils with that of severe COVID-19 patients. Transcriptome-wide analysis and candidate gene expression revealed similarities between hamster and human immune responses, particularly involving monocytes and neutrophils. Disease-related pathways across species highlighted interferon responses and inhibition of viral replication. Our structured neural network-supported workflow can be applied to other diseases, enhancing the identification of animal models with shared pathomechanisms.

 

MissingLinkText

Publication History

Article published online:
18 March 2025