Cell-cell communication analysis was performed using the CellChat R package (v1.6.1) (
Jin et al., 2021). A Seurat object containing annotated thymic epithelial cells and T-cell subsets served as input. To enhance the accuracy of ligand–receptor pairing, goose genes were mapped to their human orthologs based on one-to-one orthologous relationships identified using biomaRt, yielding 11,996 conserved orthologous gene pairs (
Table S2). CellChat objects were constructed from the normalized expression matrix and cell-type annotations, and the human CellChat ligand–receptor database was applied for downstream analysis.
Communication inference followed the standard CellChat workflow, which included identification of overexpressed genes and ligand–receptor interactions, calculation of communication probabilities, filtering of low-abundance interactions, and pathway-level aggregation. Communication networks were visualized using circle plots, bubble plots, signaling role heatmaps, and pattern-based decomposition analyses.
Pathway activity was inferred using decoupleR package v2.9.7
80 with PROGENy, a curated collection of 14 signaling pathways and their downstream target genes, with weighted gene-pathway interactions.
48 The resulting gene-level Wald test statistics were used as input to estimate PROGENy pathway activities. Inference was restricted to the top 500 target genes per pathway, ranked by PROGENy
p-value, and pathway activity scores were obtained by fitting a multivariate linear model (using
run_mlm function in decoupleR), for pathways with at least 5 mapped target genes. For visualization, pathway scores were zero-centered and scaled across groups (timepoints/cohorts) within each pathway, tissue, and immune cell type to preserve directionality while normalizing magnitude for cross-group comparisons.
Pathway activity inference. Pathway activities were inferred using the decoupleR package (v2.9.780) with the PROGENy resource, a curated collection of 14 signaling pathways and their downstream target genes with weighted gene–pathway interactions. Analyses used pre-built PROGENy networks restricted to the top 500 target genes per pathway (ranked by PROGENy p-value). The standardized expression matrix of the RNA assay (Seurat layer = "data") was used as input, and pathway activities were estimated by fitting a multivariate linear model (run_mlm in decoupleR), retaining only pathways with at least 5 mapped target genes (minsize = 5). The resulting cell-level pathway activity scores were stored as a Seurat assay and then zero-centered and scaled using ScaleData within the groups defined by group.by, thereby normalizing the magnitude while preserving directionality for cross-group comparison. Mean pathway activity was subsequently computed per group, and heatmaps were generated with pheatmap.