scMagnify infers a likely sequence of gene regulation from single-cell snapshots
scMagnify infers a likely sequence of gene regulation from single-cell snapshots
A paper published on October 17 in Nature Communications describes scMagnify, a program that uses single-cell measurements to infer a likely sequence of events that change gene activity. Each measurement captures a separate snapshot of a cell, so the order of transitions must be inferred from similarities between cell states.
One cell may already be changing its gene expression program while another still retains its previous state. scMagnify arranges similar states along pseudotime, a computationally inferred sequence of transitions between cell states. Along this sequence, the model tests whether the earlier activity of one gene improves predictions of another gene’s activity at later stages.
The model combines three types of data. Likely transitions between cells establish the order of states. Chromatin accessibility data indicate which regions of DNA proteins can access, narrowing the set of possible regulatory links. The program then compares the earlier activity of a transcription factor, a protein that switches genes on or off, with the subsequent activity of a target gene.
The model tests several possible time lags: a signal may appear at an adjacent point or farther along the inferred trajectory. It then groups the links into modules, sets of factors and target genes that are active at particular stages of a cell’s transition.
The authors evaluated the method in stages. Artificial data with a known gene network allow the inferred links to be compared with the original links. They then applied scMagnify to two datasets on human blood cell formation and to the development of the endocrine pancreas in mice, checked the results against known regulators, and assessed robustness through repeated runs, cell subsampling, and random removal of some RNA measurements.
In human kidney cortex data, the model proposed a chain of events linking signals from cells in the local environment, through receptors on the surface of epithelial cells, to factors that change the state of those epithelial cells after injury. This list helps researchers select signal and gene pairs for subsequent experimental testing.
The authors have made the scMagnify code publicly available. The method gives researchers a way to formulate testable hypotheses about which regulatory signals may have preceded a change in cell state.