GeneSpectra: a method for context-aware comparison of cell type gene expression across species

Abstract

Computational comparison of single cell expression profiles cross-species uncovers functional similarities and differences between cell types. Importantly, it offers the potential to refine evolutionary relationships based on gene expression. Current analysis strategies are limited by the strong hypothesis of ortholog conjecture, which implies that orthologs have similar cell type expression patterns. They also lose expression information from non-orthologs, making them inapplicable in practice for large evolutionary distances. To address these limitations, we devised a novel analytical framework, GeneSpectra, to robustly classify genes by their expression specificity and distribution across cell types. This framework allows for the generalization of the ortholog conjecture by evaluating the degree of ortholog class conservation. We utilise different gene classes to decode species effects on cross-species transcriptomics space and compare sequence conservation with expression specificity similarity across different types of orthologs. We develop contextualised cell type similarity measurements while considering species-unique genes and non-one-to-one orthologs. Finally, we consolidate gene classification results into a knowledge graph, GeneSpectraKG, allowing a hierarchical depiction of cell types and orthologous groups, while continuously integrating new data.