Abstract
Dissecting the cellular heterogeneity embedded in single-cell transcriptomic data is challenging. Although many methods and approaches exist, identifying cell states and their underlying topology is still a major challenge. Here, we introduce the concept of multiresolution cell-state decomposition as a practical approach to simultaneously capture both fine- and coarse-grain patterns of variability. We implement this concept in ACTIONet, a comprehensive framework that combines archetypal analysis and manifold learning to provide a ready-to-use analytical approach for multiresolution single-cell state characterization. ACTIONet provides a robust, reproducible, and highly interpretable single-cell analysis platform that couples dominant pattern discovery with a corresponding structural representation of the cell state landscape. Using multiple synthetic and real data sets, we demonstrate ACTIONet’s superior performance relative to existing alternatives. We use ACTIONet to integrate and annotate cells across three human cortex data sets. Through integrative comparative analysis, we define a consensus vocabulary and a consistent set of gene signatures discriminating against the transcriptomic cell types and subtypes of the human prefrontal cortex.
Highlights
Dissecting the cellular heterogeneity embedded in single-cell transcriptomic data is challenging
Unlike state-of-the-art clustering- or decomposition-based methods, ACTIONet implements the concept of multiresolution decomposition, an approach that enables data-driven identification of both coarseand fine-grained transcriptional patterns defining discrete and continuous cell states
ACTIONet uses a modified version of archetypal analysis (ACTION) to learn dominant transcriptional patterns representative of transcriptional cell types and states, and manifold learning to construct a structural representation of the cell-state space
Summary
Dissecting the cellular heterogeneity embedded in single-cell transcriptomic data is challenging. A priori selection of the number of underlying dominant patterns to be learned is required but not trivial to determine, nor is the certainty of robustness and reproducibility of the patterns To simultaneously address these limitations and provide a systematic way to learn cell-type discriminatory and shared dominant transcriptional patterns, here we present a computational framework built upon the concept of multiresolution cell-state decomposition, an approach that systematically prunes and unifies informative patterns identified at different levels of resolution. The outcome of this method is a nonredundant, multiresolution set of cell states whose relative contribution optimally represents the heterogeneity of the entire single-cell transcriptomic data set. This is achieved by introducing a multilevel decomposition and a multiresolution cell-state discovery approach that circumvents technical problems associated with transcriptomic decomposition, while accounting for a potential intrinsic biological property inherent to single-cell data: the existence of multiple meaningful levels of resolution that prohibit the specification of a single optimal number of clusters/ components to partition the data
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
