A novel generative adversarial networks modelling for the class imbalance problem in high dimensional omics data

Area of Science:

• Bioinformatics
• Machine Learning
• Genomics

Background:

• Class imbalance is a significant challenge in high-throughput omics analyses, leading to biased machine learning classifiers.
• Traditional oversampling methods like SMOTE and random oversampling can introduce inaccuracies, especially with high-dimensional and noisy healthcare data.

Purpose of the Study:

• To propose a novel generative adversarial network (GAN)-based method for generating synthetic samples from small, high-dimensional omics datasets.
• To improve upon existing generative approaches for addressing class imbalance in machine learning.

Main Methods:

• A generative adversarial network (GAN) model was developed to create synthetic data points for the minority class.
• The proposed GAN-based oversampling method was compared against Synthetic Minority Oversampling Technique (SMOTE) and Random Oversampling (RO).
• Classifiers were trained on data balanced using each oversampling technique to validate the generative methods.

Main Results:

• The GAN-based method demonstrated potential for generating synthetic samples in high-dimensional omics data.
• Performance evaluation through classifier training indicated improvements compared to naive oversampling techniques.
• The study highlights the effectiveness of generative models in mitigating class imbalance bias.

Conclusions:

• Generative adversarial networks offer a promising approach to address class imbalance in high-throughput omics data.
• This method provides a more robust alternative to traditional oversampling techniques, particularly for noisy, high-dimensional datasets.
• The proposed GAN-based strategy enhances machine learning classifier generalization in omics studies.