VARION: A Network Propagation Framework for Individual Patient Somatic Mutation Interpretation in Cancer Molecular Subtyping
Accurate molecular subtyping of individual cancer patients from somatic mutation data remains a challenge in precision oncology research. Existing network-based stratification (NBS) methods treat all mutations equivalently, require full-cohort batch processing, and do not demonstrate generalization to independent datasets without retraining. To address this, we present variant interpretation via…
Molecular subtyping of individual cancer patients from somatic mutation data is a significant hurdle in precision oncology research. Traditional network-based stratification (NBS) methods overlook the differences among mutations, necessitate processing the entire cohort in one go, and fail to generalize well to new datasets without additional training.
To tackle these limitations, researchers have introduced variant interpretation via the adaptive network pRopagatION (VARION). This novel framework combines population-level variant constraint scoring with protein-protein interaction (PPI) network topology.
At the heart of VARION lies the Adaptive Topology-aware Random Walk with Restart (ATR-RWR) algorithm. This algorithm assigns a weight to each mutated gene using the formula {varphi}g = {sqrt}(GIS(g) x {rho}topo(g)). The Gene Intolerance Score (GIS) quantifies population-level functional constraint, which is then propagated through a shared PPI network. By employing cosine similarity to TCGA-derived reference centroids, VARION can assign subtypes to individual patients in real-time.
In a comprehensive evaluation across ten TCGA cancer cohorts (comprising 2,417 patients), VARION demonstrated impressive accuracy: 77.7% for ovarian cancer (OV), 69.5% for glioblastoma (GBM), 90.2% for cholangiocarcinoma (CHOL), and 75.4% for gastric cancer (STAD). To assess the framework's effectiveness further, researchers conducted a controlled benchmark by applying two alternative clustering methods (PyNBS; a dense autoencoder) to the same ATR-RWR propagation matrices.
The results revealed no significant driver enrichment for PyNBS (OR = 1.79, non-significant) and PyNBS (OR = 1.52, non-significant) compared to VARION (OR = 144.29, highly significant, p = 1.77x10^-12). This clear distinction confirmed that the GIS-weighted centroid architecture, not mere propagation, is the driving force behind VARION's exceptional performance.
To ensure the framework's robustness, researchers also tested VARION's generalization capabilities in two independent cohorts (ICGC CCA, n = 396; PCAWG, n = 110). The results were striking: OR = {infty} (p < 0.001) for ICGC CCA and OR = {infty} (p < 0.001) for PCAWG, indicating that VARION can effectively classify cancer patients without the need for retraining.
These findings underscore the potential of VARION as a powerful tool for precise cancer diagnosis and classification, paving the way for more effective and personalized cancer treatments.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
