Single-cell RNA sequencing has given scientists the ability to study the activity of thousands of individual cells at once. In cancer research, this technology helps reveal how cells differ from one another and how tumors evolve over time. However, understanding the evolutionary relationships between cells remains challenging because gene expression data alone do not directly reveal how cells are related.

Researchers from ETH Zurich have developed a new computational method called SCITE-RNA that reconstructs evolutionary trees using single-cell RNA sequencing data. The approach analyzes single-nucleotide variants detected in RNA sequencing reads and uses them to infer how individual cells are related through their ancestry.

Phylogenetic trees are commonly used to describe evolutionary relationships. In cancer, these trees can reveal how tumor cells diversify and accumulate mutations over time. Existing approaches often struggle with the noisy and incomplete nature of single-cell RNA sequencing data, making it difficult to accurately reconstruct these relationships.

SCITE-RNA addresses this problem by using a maximum-likelihood search strategy that alternates between two different representations, one based on cell lineage trees and another based on mutation trees. By switching between these views, the algorithm avoids becoming trapped in local solutions and improves the quality of the inferred evolutionary history.

SCITE-RNA method overview

First, RNA of single cells is sequenced, followed by the selection of potentially mutated loci. We then obtain a likelihood for the read counts of each cell–mutation pair given the inferred genotypes and hence a joint likelihood for any given cell lineage tree or mutation tree. To find the maximum likelihood tree, we explore the tree space greedily with simple tree moves. Once a set of moves is completed, the tree is converted into the alternative space, where the exploration continues. This process is repeated until the tree hits a local optimum in both spaces. Using the genotypes inferred by the final tree, we optimize the model parameters to better fit a specific dataset and continue with the second round of tree optimization

The researchers first tested the method using simulated datasets. SCITE-RNA consistently performed better than existing approaches in reconstructing phylogenetic relationships. The team then applied the method to cancer single-cell RNA sequencing datasets.

The analysis allowed the researchers to connect cellular evolution with gene expression patterns. This combination provides insights not only into how cancer cells are related, but also into how changes in gene activity accompany tumor development.

Tools such as SCITE-RNA could help researchers better understand tumor heterogeneity and identify how different cell populations emerge during disease progression. As single-cell technologies continue to improve, methods that combine genetic variation with gene expression information may provide a more complete picture of cellular evolution.

Availability – The code is published on GitHub https://github.com/cbg-ethz/SCITE-RNA under the GNU General Public License version 3.