Abstract
Single-cell RNA sequencing (scRNA-Seq) enables advanced analysis of cellular heterogeneity; however, due to data sparsity and noise, analyzing single genes can be challenging. While pathway activity scoring offers a robust alternative to standard single-gene analysis, most available enrichment methods were developed for bulk RNA-Seq or microarray data, so they may fail in addressing the variability and dropout typical of single-cell experiments. Existing approaches primarily focus on comparing pathway activity across predefined groups. Thus, there is a gap in solutions for clustering single-pathway activity vectors, which helps detect subpopulations. Consequently, this limits the functional interpretation and the discovery of meaningful data heterogeneity. We introduce FUNCellA, a framework for estimating relative pathway activity scores using unsupervised tools such as k-means and Gaussian mixture modeling. Integrating 7 single-sample enrichment algorithms with novel relative activation thresholding methods, FUNCellA identifies active, inactive, and intermediate cellular states. Benchmarking across scRNA-Seq datasets, both real and simulated one, as well as bulk and microarray transcriptomes, the proposed solution yields outperforming results in relative pathway activation detection compared to existing tools. Finally, FUNCellA enables functional cell classification beyond marker-based clustering, uncovering heterogeneity such as sub-activation states and disease-specific responses.