NonLoss: a novel analytical method for differential biological module identification from single-cell transcriptome

Background: The identification of disease-related biological modules plays an important role in our understanding of the process of diseases. Although single-cell RNA sequencing (scRNA-seq) provides high-resolution transcriptome data that can potentially characterize subtle gene expression changes within cells, the susceptibility of the gene expression information to the influence of individual genes also makes it difficult to distinguish the biological module. Methods: To quantify gene expression information for biological function modules, we adopted the method based on Shannon's entropy and Spearman rank correlation analysis. The ingenious combination of these two methods enables the variation analysis of the former and the consistency analysis of the latter to make a more robust biological function analysis tool. Results: We developed a computational analytical method and desktop application called NonLoss to analyze scRNA-seq data more robustly and to extract real biological differences between cell populations. The method derives its power by handling expression level data from all genes annotated to a specific function module, both for dimensionality reduction and reliability of function identification, avoiding random disturbance of individual genes. NonLoss can in principle be used to assess changes of function modules and identify vital functions simultaneously. Furthermore, specific genes contributing to important functions, even those with subtle expression changes, can be identified. The results demonstrated that NonLoss yields biologically significant insights into 3 different applications. Conclusions: NonLoss was developed with a user-friendly graphical user interface, and it could identify the module of biologically relevant expression changes at a single-cell resolution.

Automated summary

The study introduced a method to robustly analyze single-cell RNA sequencing data by considering all genes annotated to specific functional modules, achieving reliable function identification and dimensionality reduction while avoiding random disturbance of individual genes.