Comparative single-cell trajectory network enrichment identifies pseudo-temporal systems biology patterns in hematopoiesis and CD8 T-cell development

Abstract

Single cell transcriptomics (scRNA-seq) technologies allow for investigating cellular processes on an unprecedented resolution. While software packages for scRNA-seq raw data analysis exist, no method for the extraction of systems biology signatures that drive different pseudo-time trajectories exists.Hence, pseudo-temporal molecular sub-network expression profiles remain undetermined, thus, hampering our understanding of the molecular control of cellular development on a single cell resolution. We have developed Scellnetor, the first network-constraint time-series clustering algorithm implemented as interactive webtool to identify modules of genes connected in a molecular interaction network that show differentiating temporal expression patterns. Scellnetor allows selecting two differentiation courses or two developmental trajectories for comparison on a systems biology level.Scellnetor identifies mechanisms driving hematopoiesis in mouse and mechanistically interpretable subnetworks driving dysfunctional CD8 T-cell development in chronic infections. Scellnetor is the first method to allow for single cell trajectory network enrichment for systems level hypotheses generation, thus lifting scRNA-seq data analysis to a systems biology level. It is available as an interactive online tool at https://exbio.wzw.tum.de/scellnetor/. Introduction Single-cell RNA sequencing (scRNA-seq) allows researchers to perform cellular developmental studies with a hitherto unseen fine granularity. Single-cell transcriptomes have paved the way for novel discoveries in various biomedical fields by improving the understanding of how transcriptional profiles relate to cell phenotypes. A range of algorithms have been invented for clustering of scRNA-seq data and for inferring differentiation trajectories 1-3 . Clustering assumes that single cells can be divided into distinct groups whereas trajectory inference aims to arrange cells such that continuous phenotypes can be traced on a low dimensional cell map 4 . Important examples of the latter include diffusion maps 5,6 and pseudotemporal ordering of single cells 3,7 . Both algorithms seek to position single cells such that their coordinates reflect their developmental statuses in relation to the other cells. Additionally, several software packages have been developed for the entire analysis pipeline, from pre-processing to clustering and identification of differentially expressed genes. Scanpy 8 , Seurat 9 and SINCERA 10 are examples of such software packages. Though scRNA-seq data is still challenged by noise 11 , combinations of different tools and algorithms have helped to unravel hidden inter-cellular mechanisms and shed light on unknown cellular paths of differentiation and disease progression 12-14 .Typical computational analyses of single cell gene expression data involve a pre-processing step, where, e.g., cells with high levels of mitochondrial DNA and few expressed genes are removed 2,15-18 . This is followed by a clustering of single-cells' transcriptional profiles and/or infere...