MK-TGAN: A Novel Knowledge-Guided Generative Model for Synthetic Transcriptomic Data

A new preprint on arXiv (2608.13256) presents a comparative analysis of generative models for transcriptomic data, focusing on incorporating prior biological knowledge via gene graphs. The study introduces MK-TGAN, an innovative multi-kernel, Graph Neural Network-based Generative Adversarial Network, which outperforms other methods in generating synthetic data that captures real-world gene patterns. The research addresses challenges such as data imbalances, biases, and ethical or legal constraints that limit access to high-quality biomedical datasets. By leveraging prior knowledge graphs, MK-TGAN ensures the synthetic data remains useful for downstream tasks. The paper benchmarks three variants of GANs, with MK-TGAN standing out for its realism and utility. This work is significant for biomedical research, as it offers a solution to data scarcity and quality issues, potentially enabling more robust data-intensive tools. The preprint was announced as a cross-type submission and is available on arXiv.