Integrating zero-inflation correction and transcriptional kinetics for single-cell transcriptomic analysis

Single-cell transcriptomic data exhibit pervasive zero inflation, while traditional models either neglect this issue or fail to capture transcriptional burst-driven bimodality, hindering accurate gene regulatory studies. This study developed a zero-inflated telegraph model that integrates technical zero correction with the stochastic gene state-switching dynamics of the classical telegraph model. Systematic validation was conducted using synthetic data, human scRNA-seq data from lupus and breast cancer patients, and mouse embryonic stem cell scRNA-seq data. The model showed superior performance: it accurately fits mRNA distributions (including bimodal patterns), reliably estimates effective transcriptional burst parameters while preventing overfitting, thus enables correction of traditional models' regulatory inference bias. It also outperforms conventional approaches in detecting differentially expressed genes, with notable advantages in small samples, and identifies unique disease-related genes (e.g., LDLR, GZMB for lupus, FAIM2, VDR for breast cancer). This biologically interpretable and robust tool advances single-cell transcriptomic analysis.