Bi-level Graph Learning Unveils Prognosis-Relevant Tumor Microenvironment Patterns in Breast Multiplexed Digital
Zhenzhen Wang1,2, Cesar A Santa-Maria3,4, Aleksander S Popel1
1Department of Biomedical Engineering, Johns Hopkins University.
Biorxiv : the Preprint Server for Biology
|May 7, 2024
Summary
This study introduces an interpretable deep learning method to identify tumor microenvironment patterns for cancer patient risk stratification. These novel biomarkers offer new prognostic insights for breast cancer and potentially other cancer types.
Area of Science:
- Oncology
- Computational Biology
- Bioinformatics
Background:
- The tumor microenvironment (TME) is crucial for cancer progression and patient prognosis.
- Current deep learning models for TME analysis lack interpretability, limiting biomarker discovery.
- Generalizable, data-driven biomarkers for TME are needed for improved prognostic tools.
Purpose of the Study:
- To develop a data-driven and interpretable approach for identifying TME cell organization patterns linked to patient prognosis.
- To create a novel risk stratification system based on TME characteristics.
- To provide new insights into the prognostic implications of the breast tumor microenvironment.
Main Methods:
- Constructed a bi-level graph model integrating cellular and population graphs.
- Utilized a soft Weisfeiler-Lehman subtree kernel to capture inter-patient similarities.
- Applied the methodology to a breast cancer cohort for pattern identification and risk stratification.
Main Results:
- Identified specific tumor microenvironment patterns associated with distinct patient prognoses.
- Developed a risk stratification system that complements existing standard systems.
- Validated the findings in two independent breast cancer cohorts, demonstrating generalizability.
Conclusions:
- The developed interpretable approach effectively identifies prognostic TME patterns.
- The TME-based risk stratification system offers complementary prognostic information.
- This methodology holds potential for application across various cancer types to understand cellular organization and patient outcomes.
