NISP: State Transition-Driven Nonlinear Imputation for Dropout Recovery in scRNA-Seq Data

Single-cell RNA sequencing (scRNA-seq) has emerged as a transformative omics technology for cell type identification in cancer diagnostics, enabling high-throughput parallel generation of cellular-resolution data that revolutionizes precision medicine. However, scRNA-seq data are frequently compromised by prevalent drop-out events due to limitations in sample quality and technical bottlenecks, which cause a large number of the results lost. To address this challenge, we propose a nonlinear imputation via state transition process (NISP) method for the diffusion and imputation of missing values in single-cell sequencing data. Our results demonstrate that the NISP framework effectively preserves nonlinear characteristics inherent to biological state transitions, enabling to recover more than 50% of missing values and remove more than 98% of noise. Therefore, NISP exhibits superior sensitivity in missing value imputation and significantly enhances the structural clarity of post-imputation datasets. Finally, validation using spatiotemporal transcriptomic arrays derived from colorectal cancer organoids further corroborates the capability of NISP to accurately capture the intrinsic manifold structure of cellular states. The result shows the significant potential of NISP in biological applications, notably its pivotal role in elucidating the mechanisms driving tumorigenesis and cancer progression.