Chimera: Cross-Species Single-Cell Generative Modeling of Expression Deployment Relative to Protein Conservation

Chimera aligns single-cell atlases across species and finds that ortholog expression deployment evidence ({Delta}[a]) can diverge from protein-sequence conservation, including in held-out species.

• Why it matters: Protein identity predicts what genes could do, but it does not measure how genes are deployed in a species’ cells. Chimera fills this gap by quantifying deployment relative to what protein conservation would predict.
• What they did: The method fits one generative model to cross-species single-cell atlases, producing shared expression-pattern evidence for one-to-one orthologs and a deviation coordinate {Delta}[a]. It reports that across eight tissues, protein identity tracks dN/dS while {Delta}[a] tracks neither.
• The result: In eight held-out species, spermatocyte genes like ZPBP2 keep stage patterns on less conserved proteins, while conserved hepatocyte enzymes like ALAS1 share patterns but not levels. More broadly, specialized-cell genes sit above protein predictions in 23 of 24 cohorts, and independent data support the count evidence in liver and testis, including a pre-specified liver test for {Delta}[a].

AI summary of the abstract · may contain errors · not medical advice · read the paper.