TL;DR
This paper introduces PESSA, a novel web-based tool that leverages pathway activation levels (gene set enrichment scores) as biomarkers for predicting cancer patient survival. By integrating data from multiple large public databases and employing the ssGSEA method, PESSA offers a user-friendly platform for comprehensive survival analysis, addressing a gap in existing tools that often focus on single genes or limited datasets.
What Problem Does This Paper Address?
Identifying reliable biomarkers for cancer prognosis is crucial for personalized treatment and improved patient outcomes. While individual gene expression has been extensively studied, researchers are increasingly recognizing the importance of biological pathways. The collective activation or deactivation of genes within a pathway can provide a more robust and less volatile indicator of a tumor's behavior than single genes. However, existing web tools for survival analysis often have limitations: they may cover only a subset of available cancer data, require significant bioinformatics expertise, or fail to fully utilize the power of pathway-level analysis. This leaves a need for an accessible, comprehensive tool that can leverage pathway activation scores for prognostic predictions across a wide range of cancers.
Methods
PESSA's core methodology revolves around the single-sample gene set enrichment analysis (ssGSEA) algorithm. This approach allows for the calculation of gene set activation levels for individual samples within a dataset. The authors utilized the Molecular Signatures Database (MSigDB), a comprehensive resource containing over 10,000 gene sets, to represent various biological pathways and processes.
For survival analysis, PESSA employs two main strategies:
- Dichotomous Grouping: Gene sets are grouped into "high" and "low" activation categories based on either the median or an optimal cut-off value. The optimal cut-off is determined using the
surv_cutpointfunction from thesurvminerR package, aiming to maximize the statistical significance of the survival difference between groups. - Continuous Variable Analysis: The ssGSEA scores are also treated as continuous variables.
Both grouping methods are then used in conjunction with Kaplan-Meier (KM) survival analyses (using the survival package) and Cox proportional hazards regression models (using the coxph function) to assess the prognostic value of the gene sets. The tool provides interactive visualizations of KM curves and detailed statistical outputs, including hazard ratios and p-values. The entire platform is built using the R Shiny framework, making it accessible via a web browser.
Datasets
PESSA integrates a substantial amount of data from several major sources:
- Gene Expression Omnibus (GEO): 180 datasets were incorporated.
- The Cancer Genome Atlas (TCGA): 33 datasets were included.
- The European Genome-phenome Archive (EGA) and supplementary tables of articles: 25 datasets related to immune checkpoint inhibitor therapy were added.
In total, PESSA covers 51 cancer types and 13 different survival outcome types. The analysis utilizes 13,434 tumor-related gene sets sourced from MSigDB, encompassing hallmark gene sets, conventional biological pathways (C2CP), and GO-based tumor-related gene sets (C5GO).
Key Insights and Ideas
- Pathway Activation as a Prognostic Biomarker: The central idea is that the collective activity of genes within a pathway is a more powerful and stable predictor of patient survival than individual gene expression. This moves beyond the limitations of single-gene markers.
- ssGSEA for Individual Sample Analysis: The use of ssGSEA is key, as it allows for the calculation of pathway activation scores for each patient sample, enabling personalized survival predictions.
- Comprehensive Data Integration: By pooling data from GEO, TCGA, and EGA, PESSA offers a pan-cancer perspective, allowing for broader discoveries and comparisons across different tumor types.
- User-Friendly Web Tool: The development of PESSA as an R Shiny application democratizes access to complex survival analysis, making it available to researchers without extensive bioinformatics backgrounds.
Key Results
PESSA successfully demonstrates the utility of pathway activation levels as prognostic markers. For instance, the paper highlights an example where the "TGF-beta Receptor Signaling in Emt Epithelial to Mesenchymal Transition" gene set was used to analyze TCGA-BLCA data. Patients with high activation of this pathway showed significantly shorter overall survival, consistent with previous findings. Another example showed that the "DNA Damage Repair (DDR)" pathway activation was associated with survival outcomes in patients treated with PD1 inhibitors for bladder cancer and melanoma. The tool provides interactive Kaplan-Meier plots and detailed statistical tables for survival analysis, facilitating the exploration of these relationships.
Limitations and Future Work
The authors acknowledge several limitations and outline future directions:
- Data Scope: While PESSA incorporates a large dataset, it acknowledges that it does not include all available transcriptome data (e.g., from SRA, EBI, or published articles not yet integrated). Future work will involve regularly updating and expanding the data repository.
- User Data Upload: Currently, PESSA does not allow users to upload their own custom datasets. This is a significant limitation for researchers with unique samples. The authors suggest users can submit data for potential inclusion.
- Custom Cut-off Values: PESSA currently relies on pre-processed median or optimal cut-off values and does not support user-defined cut-offs, limiting personalized analysis.
- Limited Clinical Information: For some datasets, information on tumor grading, treatment measures, and histology is missing, preventing subgroup analyses based on these factors. The authors aim to address this by incorporating more detailed treatment information.
- Future Tool Development: The team plans to develop additional web tools for tumor survival analysis, including those focusing on proteomics and epigenomics, to provide a more comprehensive exploration of tumor biology.
Summary updated