Orthogonal Shared Basis Factorization: Cross-species gene expression analysis using a common expression subspace

Abstract

One of the main challenges in analyzing gene expression profiles across species is the dependence on determining corresponding genes between species. Homology-based approaches fail to account for the contribution of non-homologous genes to the phenotype, genes functional divergence, and rewiring of pathways. Homology-independent methods based on joint matrix factorization provide a potential solution, but biological interpretations with existing approaches are difficult. We developed a novel joint matrix factorization method that we call the orthogonal shared basis factorization (OSBF) to compare functionally similar phenotypes across species. OSBF utilizes a similar correlation structure within individual datasets to estimate interpretable matrix factors. This homology-independent approach places cellular phenotypes in a common coordinate system that can summarize gene expression patterns shared by different organisms and quantifies the role of all genes in the phenotype independent of their homology relationships and annotation. OSBF is available on GitHub.