Mice are widely used as biomedical research models, yet results from mouse gene studies often differ from equivalent studies in humans despite similarities in DNA sequences. Here we show that gene expression patterns provide additional insight into human–mouse gene relationships. Using the Transformer-based GeneRAIN model on 777 K human and mouse bulk RNA sequencing samples, we generate RNA-based representations of genes and compare homologous genes across species. We identify 2,407 human–mouse homologous genes with high DNA sequence similarity but distinct RNA characteristics, which are more likely to have different disease or phenotype associations. We also identify 3,070 genes with low similarity at both DNA and RNA levels, suggesting a high risk of cross-species discrepancies. Our approach also identifies long noncoding RNA homologs based on cross-species RNA conservation. These results provide a resource for evaluating the suitability of mouse models for studying specific human genes. Deep learning of 777 K human and mouse transcriptomes reveals RNA-level gene similarities beyond DNA sequence, identifying homologous genes with divergent functions and helping guide the use of mouse models.
Su et al. (Wed,) studied this question.