Answers are grounded in this paper and cite their sources.
Generative artificial intelligence (GenAI) models have become central to modern Artificial Intelligence (AI) systems and analysis in high-dimensional, low-sample-size (HDLSS) biological domains. This article presents a critical review and comparative analysis of major Gen-AI families—including Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), diffusion-based probabilis-tic models, and energy-based models—with emphasis on their behavior, stability, and generative fidelity under data-scarce conditions. We survey the literature from 2015–2025, organize the field through a taxonomy of current generative paradigms, and analyze how these models are adapted to omics settings such as 1 scRNA-seq, bulk RNA-seq, spatial transcriptomics, metabolomics, and related molecular data. A recurring finding across the literature is that most models rely on aggressive preprocessing, such as highly variable gene selection, principal component analysis, or the latent space structure of an encoder-decoder network , effectively replacing the original HDLSS problem with a lower-dimensional proxy. These models are typically applied to moderate to medium-sized data sets and currently lack validation in biological applications where sample size is significantly small, e.g., fewer than 10 replicates. We further identify the absence of embedded sparsity-aware feature selection as a major methodological gap. To complement the survey, we experimentally compare representative VAE-, GAN-, and diffusion-based models on benchmark biological datasets in the HDLSS regime. The results show that VAEs remain comparatively stable in low-sample settings, whereas GANs and diffusion models improve substantially as sample size increases. Overall, this review provides a structured perspective on genera-tive modeling under HDLSS biological constraints and highlights open directions in data-efficient, interpretable, and reproducible Gen-AI for biological discovery.
No takes yet. Share an insight, caveat, or question.
Ghosh et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: