AC-WGAN-GP for transcriptomic data augmentation: Enhancing stress classification in Synechocystis sp. PCC 6803 under...

Area of Science:

• Microbial transcriptomics
• Computational biology
• Machine learning applications in genomics

Background:

• Transcriptomic studies in Synechocystis sp. PCC 6803 face challenges like small sample sizes, class imbalance, and technical batch effects.
• These limitations hinder reliable machine learning for multi-class stress classification.

Purpose of the Study:

• To develop and evaluate a method for augmenting small, imbalanced microbial transcriptomic datasets.
• To improve the accuracy and reliability of machine learning models for classifying abiotic stress responses in Synechocystis sp. PCC 6803.

Main Methods:

• Integrated transcriptomic data from 12 studies (80 samples) and applied batch effect correction.
• Utilized an Auxiliary Classifier Wasserstein Generative Adversarial Network with Gradient Penalty (AC-WGAN-GP) for synthetic data generation.
• Evaluated data augmentation impact using classification metrics, distributional fidelity, and manifold analysis.

Main Results:

• Batch correction harmonized cross-platform data, improving sample alignment by biological condition.
• Moderate synthetic data augmentation (500 samples/class) maximized classification accuracy improvement (0.800 ± 0.028).
• Distributional fidelity and manifold alignment improved with increased synthetic data, outperforming SMOTE and conventional GANs.

Conclusions:

• Wasserstein-based conditional generative models effectively augment imbalanced microbial transcriptomic data.
• AC-WGAN-GP offers a robust approach for analyzing underrepresented stress conditions.
• Task-specific optimization is crucial for balancing predictive performance and distributional realism in data augmentation.