Applying a GAN-based classifier to improve transcriptome-based prognostication in breast cancer.

Area of Science:

• Computational biology
• Genomics
• Artificial intelligence in medicine

Background:

• Established breast cancer prognostic tests use limited transcripts and have specific approval criteria.
• Developing robust classifiers from omics data is challenging due to high dimensionality (more variables than patients).

Purpose of the Study:

• To develop a robust deep learning classifier for stratifying breast cancer patients based on full transcriptome data.
• To overcome data limitations in omics datasets using advanced generative adversarial networks.

Main Methods:

• Proposed a classifier using a data augmentation pipeline with a Wasserstein generative adversarial network (GAN) and an auxiliary classifier (T-GAN-D).
• Applied the T-GAN-D classifier to the METABRIC breast cancer cohort (1244 patients).
• Validated performance across independent, merged transcriptome datasets (METABRIC and TCGA-BRCA).

Main Results:

• The T-GAN-D classifier outperformed established breast cancer biomarkers in distinguishing low-risk from high-risk patients.
• The classifier demonstrated effectiveness across independent and merged datasets.
• Merging transcriptome data improved overall patient stratification accuracy.

Conclusions:

• Reiterative GAN-based training yields a robust classifier for full transcriptome data analysis.
• The developed classifier can accurately stratify breast cancer patients into risk groups across heterogeneous cohorts.
• This approach enhances prognostic capabilities for breast cancer management.