A topological approach for cancer subtyping from gene expression data

Area of Science:

• Computational biology
• Bioinformatics
• Cancer genomics

Background:

• Gene expression data is crucial for cancer subtyping.
• High dimensionality of omics data presents a challenge ('curse of dimensionality') for computational methods.
• Existing methods struggle to clearly distinguish cancer subtypes based on survival plots.

Purpose of the Study:

• To propose a novel framework for cancer subtyping using Topological Mapper.
• To develop a new method for defining the filter function crucial for the Mapper algorithm.
• To evaluate the survival analysis of discovered cancer subtypes and their separation.

Main Methods:

• Utilized the Topological Mapper algorithm for cancer subtyping.
• Developed a novel approach for defining the filter function for the Mapper algorithm.
• Performed survival analysis using Kaplan-Meier plots and hazard ratios to evaluate subtype separation.

Main Results:

• The proposed methodology demonstrated superior separation of cancer subtypes compared to RSC-Otrimle and SNF algorithms across five cancer datasets.
• Achieved greater minimum pairwise life expectancy differences (e.g., 425 days for kidney cancer) compared to existing methods.
• Hazard ratio analysis indicated improved performance in four out of five datasets, with reduced overlap in Kaplan-Meier plots for lung, breast, and kidney cancers.

Conclusions:

• The developed framework effectively addresses the curse of dimensionality in cancer gene expression data.
• The improved subtype separation aids in a more individualized understanding of cancer patients.
• This work supports the advancement of Precision Medicine through patient-specific cancer insights.