Deep learning assisted cancer disease prediction from gene expression data using WT-GAN

Area of Science:

• Bioinformatics
• Computational Biology
• Artificial Intelligence in Healthcare

Background:

• Deep learning (DL) adoption is growing in healthcare and drug development.
• Cancer remains a leading cause of mortality, necessitating improved diagnostic and predictive tools.
• Limited gene expression data presents a challenge for training DL models in cancer research.

Purpose of the Study:

• To address the scarcity of gene expression data for DL models in cancer diagnosis.
• To evaluate the effectiveness of data augmentation using a Wasserstein Tabular Generative Adversarial Network (WT-GAN).
• To improve the accuracy of cancer diagnosis through enhanced gene expression data analysis.

Main Methods:

• Utilized the Wasserstein Tabular Generative Adversarial Network (WT-GAN) for synthetic gene expression data augmentation.
• Employed correlation-based feature selection to identify relevant genetic characteristics.
• Trained and classified gene expression samples using Deep Feedforward Neural Network (FNN) and Machine Learning (ML) algorithms.

Main Results:

• Data augmentation with WT-GAN significantly expanded the training dataset size.
• Correlation-based feature selection identified key genetic markers.
• The augmented dataset led to improved classification performance, achieving over 97% accuracy in cancer diagnosis.

Conclusions:

• WT-GAN is an effective method for augmenting gene expression data, overcoming limitations of sample size and dimensionality.
• Enhanced data through augmentation improves the performance of DL and ML models for cancer diagnosis.
• This approach holds promise for more precise and timely cancer detection and prediction.