ssMutPA: single-sample mutation-based pathway analysis approach for cancer precision medicine

Yalan He, Jiyin Lai, Qian Wang, Bingyue Pan, Siyuan Li, Xilong Zhao, Ziyi Wang, Y. Zhang, Yujie Tang, Junwei HanGigaScience 2024 · DOI

TL;DR

This paper introduces ssMutPA, a novel computational method that analyzes cancer at the individual patient level by integrating somatic mutation data with protein-protein interaction networks. This approach allows for a more precise understanding of cancer subtypes and potential therapeutic targets, complementing existing methods that rely solely on gene expression data.

What Problem Does This Paper Address?

Precision medicine aims to tailor treatments to individual patients. A key challenge in achieving this is understanding the complex molecular underpinnings of cancer, which vary significantly between individuals and even within the same cancer type. Traditional approaches to analyzing cancer molecular data often focus on gene expression, which can be highly dynamic and influenced by various factors, leading to issues with reproducibility and clinical applicability. Furthermore, these methods often overlook the crucial role of somatic mutations, which are fundamental drivers of cancer development. While some methods attempt to analyze pathways (groups of interacting genes) to capture this complexity, they typically require large sample sizes and fail to account for individual patient mutations. This paper addresses the need for a method that can infer individualized pathway activity directly from somatic mutation data, providing a more robust and personalized approach to cancer analysis.

Methods

The core innovation of this paper is the ssMutPA (single-sample mutation-based pathway analysis) approach. It works by:

  1. Mapping Mutations to a Network: Somatic mutation data from individual patients is mapped onto a human-specific protein-protein interaction (PPI) network. This network represents how genes and their protein products interact.
  2. Local Weighted Strategy: For each mutated gene (a "seed node"), ssMutPA assesses the importance of its neighboring genes within the PPI network. This is done by considering the number of mutated genes in the neighborhood and the total number of neighbors, using a hypergeometric distribution to quantify the reinforcement of a seed node's importance by its mutated neighbors. This local weight aims to capture how mutations in a gene's immediate functional context influence its activity.
  3. Global Propagation Strategy: To account for how mutations can indirectly affect genes further away in the network, ssMutPA employs a "random walk with restart" (RWR) algorithm. This algorithm simulates a random walker moving through the network, with a probability of returning to the starting (mutated) gene. By assigning initial weights based on the local strategy, this global propagation helps to estimate the influence of mutated genes across the entire network.
  4. Pathway Enrichment Score (ssMutPES): Finally, ssMutPA calculates a single-sample mutation-based pathway enrichment score (ssMutPES) for each pathway. This score reflects the accumulated effect of mutations on the genes within a specific pathway, providing a measure of individualized pathway activity driven by mutations.

Datasets

The study utilized several datasets for its analysis:

  • The Cancer Genome Atlas (TCGA): Data from 33 cancer cohorts was used for pan-cancer analysis, including somatic mutation data and clinical information.
  • Independent Glioma Datasets: Two additional glioma datasets (Varn et al. cohort and Wang et al. cohort) from the cBioPortal were used to validate the generalizability and robustness of the ssMutPA method.
  • Immunotherapy Datasets: Two melanoma cohorts treated with immune checkpoint inhibitors (ICIs) were used to assess ssMutPA's ability to identify patient subgroups that might benefit from therapy. These included a dataset from Liu et al. (PD-1 blockade) and a dataset from Snyder et al. (CTLA-4 blockade).
  • KEGG Database: 323 pathways were downloaded from the KEGG database to represent biological pathways.
  • Human-Specific PPI Network: This network was compiled from 12 sources and filtered for high-confidence links.

Key Insights and Ideas

  • Mutation-Driven Pathway Activity is Key: The study strongly suggests that somatic mutations, not just gene expression, are critical drivers of pathway dysregulation and cancer progression. ssMutPA provides a way to quantify this mutation-driven activity at an individual level.
  • Network Topology Matters: The integration of the PPI network is crucial. ssMutPA's local and global weighted strategies leverage network structure to understand how mutations propagate their effects, going beyond simple gene-level analysis.
  • Identifying Novel Cancer Subtypes: By analyzing mutation-based pathway activity, ssMutPA can stratify patients into distinct subtypes with significantly different prognoses, offering a new lens for understanding cancer heterogeneity.
  • Cross-Cancer Common Pathways: The identification of 14 common prognostic pathways across multiple cancer types highlights conserved molecular mechanisms that could serve as broad biomarkers.
  • Predicting Immunotherapy Response: ssMutPA demonstrated potential in identifying patient subgroups likely to respond to immunotherapy, suggesting its utility in guiding treatment decisions.

Key Results

  • Pan-Cancer Stratification: ssMutPA successfully stratified patients into prognostically distinct subtypes in 32 out of 33 TCGA cancer types.
  • Prognostic Biomarkers: 14 common pathways were identified as shared prognostic biomarkers across at least 6 cancer types.
  • Glioma Subtyping: In glioma, ssMutPA identified subtypes that correlated with clinical classifications (GBM vs. LGG) and showed distinct pathway activity profiles (e.g., mutation-induced signaling pathways in one subtype vs. metabolic pathways in another).
  • Immunotherapy Response Prediction: In melanoma patients treated with ICIs, ssMutPA identified subtypes associated with significantly different overall survival and objective response rates.
  • Robustness: The method showed robustness to variations in the PPI network structure, with performance remaining high even after removing up to 20% of network edges.
  • Superior Performance: Compared to other individualized pathway analysis methods (GSVA, ssGSEA, iPath, Pathifier) that rely on gene expression, ssMutPA showed superior or comparable performance in survival prediction and subtype identification.

Limitations and Future Work

  • PPI Network Completeness: The authors acknowledge that the performance of ssMutPA may be limited by the completeness of the current PPI network. As PPI databases are updated, the method's applicability is expected to improve.
  • Data Sparseness: While ssMutPA addresses the sparseness of mutation data by using network integration, the inherent limitations of sparse data in certain analyses might still exist.
  • Future Work: The authors suggest that ssMutPA can be further enhanced by integrating other omics data (e.g., copy number variations, methylation) and by exploring its application in a wider range of cancer types and therapeutic contexts. The development of ssMutPA as an R package aims to facilitate its broader adoption and further research.

Summary updated

Abstract

We provided evidence that ssMutPA could infer mutation-based individualized pathway activity profiles and complement the current individualized pathway analysis approaches focused on gene expression data, which may offer the potential for the development of precision medicine. ssMutPA is available at https://CRAN.R-project.org/package=ssMutPA.
BibTeX
@inproceedings{he2024ssmutpa,
  author = {Yalan He and Jiyin Lai and Qian Wang and Bingyue Pan and Siyuan Li and Xilong Zhao and Ziyi Wang and Y. Zhang and Yujie Tang and Junwei Han},
  title = {{ssMutPA: single-sample mutation-based pathway analysis approach for cancer precision medicine}},
  booktitle = {GigaScience},
  year = {2024},
  doi = {10.1093/gigascience/giae105},
}
Figure