Diffusion on PCA-UMAP Manifold: The Impact of Data Structure Preservation to Denoise High-Dimensional Single-Cell RNA Sequencing Data

Simple Summary In scRNA-seq analysis, diffusion-based approaches help identify the connections between cells, allowing us to observe the progression of individual cells as they change phenotypes within a mathematical space known as a manifold. Recently, these approaches have been used as a reference for imputation, a technique that addresses missing data, a common challenge in scRNA-seq analysis. For example, MAGIC is a popular diffusion-based imputation method, and it has shown success in uncovering gene-gene interactions related to phenotypic transitions that would not be possible without imputation. However, previous evaluations have not adequately compared the impact of different parameter settings on MAGIC, particularly over-smoothing issues. To address this, we developed sc-PHENIX, which utilizes a similar diffusion approach as MAGIC but incorporates a PCA-UMAP initialization step, whereas MAGIC only uses PCA. We compared sc-PHENIX and MAGIC in terms of imputation accuracy, visualization, biological insights, and preservation of data structure. Our findings show that sc-PHENIX outperforms MAGIC across various common parameters such as "diffusion time" (t), the number of nearest neighbors (knn), and PCA dimensions. It effectively captures and preserves the global, local, and continuous data structures, leading to more reliable imputation and potentially uncovering new biological insights in diverse datasets.Abstract Single-cell transcriptomics (scRNA-seq) is revolutionizing biological research, yet it faces challenges such as inefficient transcript capture and noise. To address these challenges, methods like neighbor averaging or graph diffusion are used. These methods often rely on k-nearest neighbor graphs from low-dimensional manifolds. However, scRNA-seq data suffer from the 'curse of dimensionality', leading to the over-smoothing of data when using imputation methods. To overcome this, sc-PHENIX employs a PCA-UMAP diffusion method, which enhances the preservation of data structures and allows for a refined use of PCA dimensions and diffusion parameters (e.g., k-nearest neighbors, exponentiation of the Markov matrix) to minimize noise introduction. This approach enables a more accurate construction of the exponentiated Markov matrix (cell neighborhood graph), surpassing methods like MAGIC. sc-PHENIX significantly mitigates over-smoothing, as validated through various scRNA-seq datasets, demonstrating improved cell phenotype representation. Applied to a multicellular tumor spheroid dataset, sc-PHENIX identified known extreme phenotype states, showcasing its effectiveness. sc-PHENIX is open-source and available for use and modification.