Fig 1.
Overview of the BranchKGN framework.
A: The framework consists of three main components: multi-omic data integration, trajectory inference, and gene-to-cell importance learning via attention-based graph modeling. B: The selected key genes are used to reconstruct differentiation trajectories and infer dynamic gene regulatory networks across pseudotime.
Table 1.
Four datasets used in this work.
Fig 2.
Accuracy of BranchKGN versus existing methods on the oRG-111 dataset.
Bar plots show the percentage of correctly identified oRG-specific genes at four ranking thresholds (Top 500, 1,000, 2,000, 5,000) for DISP, MVP, VST, scGEAToolbox, and BranchKGN. A pie chart inset indicates the proportion of known oRG-associated genes in the dataset (111 genes, 0.44%), serving as the gold-standard reference.
Fig 3.
Pseudotemporal trajectory analysis and gene expression profiling for the pbmc_unsorted_3k dataset.
A: Slingshot-inferred trajectory overlaid on cluster assignments, with six original clusters consolidated into four representative groups. B: Pseudotemporal ordering of all genes. C: Expression profile of the gene ARHGAP26 across clusters along the differentiation trajectory. D: Reconstructed trajectory using only the top 312 key genes.
Fig 4.
The expression variation of transfer genes.
A: Log2FC variation of gene sets at different branches. B: Heatmap of the expression of selected genes in differentiated clusters.
Table 2.
P-Value and FDR-adjusted p-value are used to judge the significance of Log2FC.
Fig 5.
Regulation map of partial gene expression in different branches.
This graph shows the changes in the average expression level and percentage of multiple genes in different cell branches (C1, C2, C3, C4). The dot color in the figure represents the average expression level of the gene, the darker the color, the higher the expression level, while the dot size indicates the percentage of the gene expressed in the cell, that is, how many cells express the gene.
Fig 6.
GRNS at different branch points.
A: Pre-branching network centered on MAML2, emphasizing its regulatory interactions with various genes. B: Branching point network focused on USP1. C: Post-branching phase 1 network with MAML2. D: Post-branching phase 2 network featuring MAML2, USP1, and SIPA1L1. In each panel, nodes represent genes, where key genes are highlighted at a larger size, and the thickness of edges reflects the strength of interactions. These networks reveal the dynamic rewiring of regulatory interactions at different stages of cell differentiation.
Fig 7.
Enrichment analysis by KEGG,GO and DisGeNET.
A: KEGG enrichment analysis showing significant pathways. B: KEGG pathway enriches the sankey plot. C: List of diseases associated with the analyzed gene set, ranked by their prevalence or significance.