Alternative splicing reshapes protein interaction networks through structured, directional rewiring across cancers

About

Alternative splicing generates extensive transcript diversity in cancer, but whether the resulting changes to protein-protein interaction (PPI) networks are structured across tumors or idiosyncratic to individual patients remains unresolved. Existing analyses catalog splicing events and their effects on individual interactions, but do not reconstruct patient-specific interaction networks at scale, leaving the systems-level architecture of splicing-driven rewiring uncharacterized. Here, focusing on exon skipping, the most prevalent form of alternative splicing in humans, we analyze 7,950 tumors across 28 cancer types to reconstruct patient-specific rewired PPI networks relative to matched normal tissues, distinguishing interaction gains from losses. We identify widespread and structured remodeling of protein connectivity, including a unique pattern of interaction gains in brain cancers, a conserved pan-cancer axis of interaction loss that converges on GTPase and Ras signaling, and a subset of genes that function as bidirectional switches, where rewiring direction varies across cancer types. At the patient level, tumors exhibiting extreme rewiring, especially pronounced interaction gains, are associated with significantly poorer survival, independent of tumor stage and mutational burden. These results delineate systems-level patterns through which alternative splicing reshapes protein interaction networks and contributes to tumor heterogeneity across cancers.

Getting started

conda env create -f env.yml
conda activate splitpea-pancancer

src/run_splitpea

run_splitpea.sh runs the full splitpea pipeline for one tissue. Install splitpea from ylaboratory/splitpea. TCGA and GTEx alternative splicing data can be downloaded from IRIS. COSMIC data can be downloaded from COSMIC (requires registration).

We also provide the Splitpea-generated networks used in our analysis at as a Zenodo record.

src/data_processing

Script	Description
general_analyisis.py	Loads sample-level splitpea networks; outputs edge/gene summaries, scatter data, top gene degrees, driver edge info, and heatmap data
consensus_and_consistency_analyisis.py	Aggregates edge counts across cancers and computes per-edge consistency and other metrics
mutations_mc3.R	Loads TCGA MC3 mutation data and saves per-sample mutation calls
mutations_drivers.R	Compares node degrees between mutated drivers, non-mutated drivers, and non-drivers with statistical tests
go_analysis.R	GO biological process enrichment analysis

src/plots

Script	Description
exon_skipping_dist_plot.R	Skipped exon event counts per sample across cancer types with per-cohort medians
rewiring_basic_summary_plot.R	Edge gain/loss/chaos boxplots per cancer type and scatter plots of splicing events with rewiring
stacked_driver_barplots.R	Proportions of driver-associated edge categories per cancer type
top_genes_heatmap.R	Heatmap of top genes across cancers with clustering
edge_conservation.py	Edge conservation across cancers at varying positive/negative rate thresholds
polarizing_edges.py	Clustermap of edges consistently gained or lost across cancers
nonpolarizing_edges.R	Genes from non-polarized edges (near-zero consistency) unique to each cancer
plot_outlier_analysis.py	Analysis and plots of outlier patients with extreme network rewiring