bioRxiv2025
Single-cell RNA-seq (scRNAseq) analyses performed at the cellular level aim to understand the cellular landscape of tissue sections, offer insights into rare cell-types, and identify marker genes for annotating distinct cell types. Additionally, scRNAseq analyses are widely applied to cancer research to understand tumor heterogeneity, disease progression, and resistance to therapy. Single-cell data processing is a challenging task due to its high-dimensionality, sparsity, and having imbalanced class distributions. An accurate cell-type identification is highly dependent on preprocessing and quality control steps. To address these issues, generative models have been widely used in recent years. Techniques frequently used include Variational Autoencoders (VAE), Generative Adversarial Networks (GANs), Gaussian-based methods, and, more recently, Flow-based (FB) generative models. We conducted a comparative analysis of fundamental generative models, aiming to serve as a preliminary guidance for developing novel automated scRNAseq data analysis systems. We performed a meta-analysis by integrating four datasets derived from pancreatic tissue sections. To balance class distributions, synthetic cells were generated for underrepresented cell types using VAE, GAN, Gaussian Copula, and FB models. To evaluate the performances of generative models, we performed automated cell-type classification tasks in original and dimensionality-reduced spaces in a comparative manner. We also identified differentially expressed genes for each cell type, and inferred cell-cell interactions based on ligand-receptor pairs across distinct cell-types. Among the generative models, FB consistently outperformed others across all experimental setups in cell-type classification (with an F1-score of 0.8811 precision of 0.8531 and recall of 0.8643). FB produced biologically more relevant synthetic data according to correlation structures (with a correlation discrepancy score of 0.0511) and cell-cell interactions found from synthetic cells were closely resembling those of the original data. These findings highlight the potential and promising use of FB in scRNAseq analyses. Author SummarySingle-cell RNA sequencing (scRNA-seq) analyses focus on identifying distinct cell types and marker genes. Traditional methods face challenges with high dimensionality, sparsity, and sample size imbalances among cell types, limiting automated and unbiased cell-type identification. Generative AI models address these issues by generating synthetic cells for under-represented types, preserving biological and contextual relevance, and employing embedding mechanisms to reduce sparsity and dimensionality. We compared widely used generative models (Variational Autoencoders, GANs, Gaussian Copula, and FB model) using integrated datasets. Synthetic data quality was assessed via cell type classification with Random Forest model in original and reduced feature spaces, correlation of differentially expressed genes, and ligand-receptor interaction inference. The FB model showed the highest potential for creating biologically accurate scRNA-seq profiles. We presented a guideline for automated cell-type identification systems by addressing gaps in single-cell analysis characteristics through the integration of widely used computational biology datasets and generative models (including a novel one, FB model).