A geometric approach to characterize the functional identity of single cells

Shahin Mohammadi, Vikram Ravindra, David F. Gleich, Ananth GramaNature Communications 2018 · DOI

Abstract

Single-cell transcriptomic data has the potential to radically redefine our view of cell-type identity. Cells that were previously believed to be homogeneous are now clearly distinguishable in terms of their expression phenotype. Methods for automatically characterizing the functional identity of cells, and their associated properties, can be used to uncover processes involved in lineage differentiation as well as sub-typing cancer cells. They can also be used to suggest personalized therapies based on molecular signatures associated with pathology. We develop a new method, called ACTION, to infer the functional identity of cells from their transcriptional profile, classify them based on their dominant function, and reconstruct regulatory networks that are responsible for mediating their identity. Using ACTION, we identify novel Melanoma subtypes with differential survival rates and therapeutic responses, for which we provide biomarkers along with their underlying regulatory networks.

Introduction

The advent of single-cell RNA sequencing (scRNA-seq) has provided unprecedented resolution into the "transcriptional fingerprints" of individual cells. However, extracting biological meaning from this high-dimensional data remains a significant challenge. Traditional computational approaches often struggle with "housekeeping" noise—the high expression of genes required for basic cellular survival that can mask subtle, cell-type-specific signals. Furthermore, many existing tools attempt to force cells into discrete clusters, failing to capture the functional continuum observed during processes like stem cell differentiation or cancer progression.

To address these limitations, we present ACTION (Archetypal-analysis for Cell-Type IdentificatioN), a geometric framework designed to characterize the functional identity of cells. Rather than simply grouping similar cells, ACTION identifies the primary biological "tasks" or functions that define a cell's unique state.

The ACTION Framework

The ACTION pipeline operates through a multi-stage process that integrates information theory, geometry, and network biology to provide a holistic view of cellular identity.

The ACTION pipeline: from similarity metrics to regulatory networks.

1. A Biologically Inspired Metric

Standard correlation metrics often fail in scRNA-seq due to the dominance of universally expressed genes. ACTION utilizes a specialized similarity metric that "masks" these housekeeping genes. By employing Shannon entropy, the method boosts the signal of genes that are highly specific to certain cell populations, ensuring that the functional identity is driven by informative markers rather than metabolic background noise.

2. Archetypal Analysis: Finding the "Corners" of Identity

At its core, ACTION views the gene expression profiles of all cells as a cloud of points in a high-dimensional space. The framework identifies the "corners" or extreme points of this cloud, known as archetypes.

Cells represented as a convex combination of functional archetypes (the 'corners' of the state space).

Each archetype represents a "pure" primary function or state. Every individual cell is then described as a convex combination—a weighted blend—of these archetypes. This allows for a continuous representation of cell states, where a cell might be, for example, 70% proliferative and 30% invasive.

3. Automated Model Selection and Network Reconstruction

Unlike methods that require a pre-defined number of clusters, ACTION uses a statistical monitoring technique based on archetype similarity graphs to automatically determine the optimal number of functional states. Once archetypes are established, the framework "orthogonalizes" them to identify genes unique to each state. It then reconstructs the regulatory networks, identifying the transcription factors (the "master switches") responsible for driving these specific functional gene patterns.

Validation and Clinical Discovery

The framework was validated across diverse datasets, including human brain cells and mouse cortical neurons. ACTION demonstrated superior robustness to "dropout" events—a common issue in scRNA-seq where genes go undetected—maintaining stability even when data density fell below 10%.

Case Study: Melanoma Subtyping

In a deep-dive study of 4,645 melanoma cells, ACTION uncovered novel clinical insights. While traditional methods grouped "MITF-high" tumor cells together, ACTION identified two distinct subclasses: Subclass B and Subclass C.

Analysis revealed that patients with Subclass C tumors had significantly worse survival rates compared to Subclass B. By reconstructing the underlying regulatory networks, we found that Subclass C cells are driven by the MYC oncogene. Furthermore, these cells showed increased sensitivity to drugs targeting the ERK MAPK signaling pathway, such as Refametinib and PLX-4720. This discovery provides a potential roadmap for personalized therapy based on a patient's specific cellular archetypes.

Conclusion

ACTION shifts the paradigm from simple cell clustering to a more nuanced understanding of functional identity. By identifying the geometric archetypes of cellular states, we can better understand the drivers of disease and identify targeted therapeutic interventions. Future work will explore the integration of ACTION with spatial transcriptomics to understand how a cell's functional archetype is shaped by its physical microenvironment.

Summary updated

Abstract

Single-cell transcriptomic data has the potential to radically redefine our view of cell-type identity. Cells that were previously believed to be homogeneous are now clearly distinguishable in terms of their expression phenotype. Methods for automatically characterizing the functional identity of cells, and their associated properties, can be used to uncover processes involved in lineage differentiation as well as sub-typing cancer cells. They can also be used to suggest personalized therapies based on molecular signatures associated with pathology. We develop a new method, called ACTION, to infer the functional identity of cells from their transcriptional profile, classify them based on their dominant function, and reconstruct regulatory networks that are responsible for mediating their identity. Using ACTION, we identify novel Melanoma subtypes with differential survival rates and therapeutic responses, for which we provide biomarkers along with their underlying regulatory networks.
BibTeX
@inproceedings{mohammadi2018geometric,
  author = {Shahin Mohammadi and Vikram Ravindra and David F. Gleich and Ananth Grama},
  title = {{A geometric approach to characterize the functional identity of single cells}},
  booktitle = {Nature Communications},
  year = {2018},
  doi = {10.1038/s41467-018-03933-2},
}
Figure