Weighted dimensionality reduction and robust Gaussian mixture model based cancer patient subtyping from gene...

Area of Science:

• Computational biology
• Cancer genomics
• Biostatistics

Background:

• Cancer's heterogeneity requires patient subtyping for effective treatment.
• Gene expression data presents computational challenges due to high dimensionality, noise, and outliers.
• Existing subtyping methods often yield overlapping survival plots, hindering clear distinction between cancer subtypes.

Purpose of the Study:

• To develop a robust computational framework for improved cancer patient subtyping.
• To enhance the separation and statistical significance of discovered cancer subtypes.
• To identify potential therapeutic targets through pathway analysis.

Main Methods:

• A robust clustering pipeline integrating dimensionality reduction and Gaussian mixture model-based clustering.
• Weighted gene expression matrix using median absolute deviation (MAD) before t-distributed Stochastic Neighbor Embedding (t-SNE) for dimensionality reduction.
• Quantification of subtype separation using minimum pairwise Kaplan-Meier (KM) plot separation, minimum hazard ratio, and a novel cumulative survival separation metric.

Main Results:

• The proposed method significantly improved subtype separation compared to Consensus Clustering, CNMF, Spectrum, and NEMO across five TCGA datasets.
• For Glioblastoma Multiforme (GBM), the method achieved a 61-day survival difference between subtypes, outperforming existing methods.
• Utilizing MAD scores notably enhanced subtype separation, and hazard ratio analysis confirmed superior performance.

Conclusions:

• The integration of median absolute deviation (MAD) and a robust clustering methodology effectively improves cancer subtype separation.
• Enhanced subtype separation offers greater statistical and clinical significance for understanding cancer heterogeneity.
• The findings provide a foundation for identifying more precise therapeutic targets in cancer treatment.