Extraction of biological signals by factorization enables the reliable analysis of single-cell transcriptomics

Accurately and reliably capturing actual biological signals from single-cell transcriptomics is vital for achieving legitimate scientific results, which is unfortunately hindered by the presence of various kinds of unwanted variations. Here we described a deep auto-regressive factor model known as scPhenoXMBD, demonstrated that each genes expression can be split into discrete components that represent biological signals and unwanted variations, which effectively mitigated the effects of unwanted variations in the data of single-cell sequencing. Using scPhenoXMBD, we evaluated various factors affecting IFN {beta} -stimulated immune cells and demonstrated that biological signal extraction facilitates the identification of IFN{beta}-responsive pathways and genes. Numerous experiments were conducted to show that scPhenoXMBD could be utilized successfully in enhancing cell clustering stability, obtaining identical cell populations from diverse data sources, advancing the single-cell CRISPR screening of functional elements, and minimizing the influence of inter-subject discrepancies in the cell-disease relationships. scPhenoXMBD is anticipated to be a dependable and repeatable method for the precise analysis of single-cell data.