Patient-specific data fusion defines prognostic cancer subtypes

Area of Science:

• Computational Biology and Bioinformatics
• Genomics and Transcriptomics
• Cancer Research

Background:

• Integrating diverse data types like copy number alterations and transcriptomics offers a more complete view of cancer biology.
• Cancer heterogeneity and measurement noise complicate data fusion, impacting accurate cancer subtype discovery and personalized therapy.
• Existing methods struggle to effectively handle discordant biological signals within patient samples.

Purpose of the Study:

• To develop a nonparametric Bayesian model for integrating gene expression and copy number variation data to discover prognostic cancer subtypes.
• To address challenges in data fusion, including separating concordant from discordant signals, feature selection, and estimating the number of subtypes.
• To enable patient-specific analysis by assessing signal concordance individually for each patient.

Main Methods:

• A nonparametric Bayesian model utilizing a hierarchy of Dirichlet Processes.
• Integration of gene expression and copy number variation data.
• Patient-specific assessment of signal concordance to distinguish reliable biological signals from noise.

Main Results:

• The model successfully identified an entirely new prostate cancer subtype with a significantly poorer survival outcome, missed by other analyses.
• In breast cancer, the model identified subtypes with superior prognostic value by leveraging concordant signals.
• The patient-specific approach and ability to distinguish concordant/discordant signals were crucial for these discoveries.

Conclusions:

• The proposed nonparametric Bayesian model effectively integrates multi-omics data for robust cancer subtype discovery.
• A patient-specific approach, accounting for signal concordance, is vital for uncovering clinically relevant cancer subtypes.
• This method enhances prognostic accuracy and has the potential to guide personalized cancer therapy strategies.