Figure 1
IA-SVA is a robust statistical framework to detect and estimate multiple and correlated hidden sources of variation. (A) Six-step IA-SVA procedure. IA-SVA computes the first principal component (PC1) from read counts adjusted for all known factors and tests its significance [Steps 1–3]. If significant, IA-SVA uses this PC1 to infer a set of genes associated with the hidden factor [Steps 4–5] and obtain a surrogate variable (SV) to represent the hidden factor using these genes [Step 6]. (B) IA-SVA uses single-cell gene expression data matrix and known factors to detect hidden sources of variation (e.g., cell contamination, cell-cycle status, and cell type). If these factors match to a biological variable of interest (e.g., cell type assignment), genes highly correlated with the factor can be detected and used in downstream analyses (e.g., data visualization).
