AEGAN-Pathifier: a data augmentation method to improve cancer classification for imbalanced gene expression data.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Cancer classification faces challenges due to high-dimensional, imbalanced, and noisy gene expression data.
• Acquiring patient samples is costly and resource-intensive, often leading to data imbalances.
• High dimensionality and limited samples can cause overfitting and the curse of dimensionality.

Purpose of the Study:

• To develop an effective deep learning method for addressing challenges in cancer classification with imbalanced gene expression data.
• To improve the performance of cancer classification models by generating synthetic samples for the minority class.

Main Methods:

• Proposed AEGAN (AutoEncoder and Generative Adversarial Network) to generate synthetic samples for imbalanced gene expression data.
• Integrated pathway prior knowledge using the pathifier algorithm to calculate pathway scores.
• Developed AEGAN-Pathifier, a data augmentation approach combining deep learning with pathway information for dimensionality reduction and biological functionality preservation.

Main Results:

• AEGAN effectively generates synthetic samples for minority classes in imbalanced gene expression datasets.
• AEGAN-Pathifier demonstrated improved performance in cancer classification tasks.
• Validation with various classifiers confirmed enhanced classifier performance and good generalization capability.

Conclusions:

• AEGAN-Pathifier significantly improves performance on imbalanced datasets like GSE25066, GSE20194, BRCA, and Liver24.
• The method shows robust generalization capabilities across different classifiers.
• This approach offers a promising solution for cancer classification using gene expression data.