Extended Data Fig. 5: Consensus non-negative matrix factorization (cNMF) and gene expression program (GEP) quality control. | Nature Methods

Extended Data Fig. 5: Consensus non-negative matrix factorization (cNMF) and gene expression program (GEP) quality control.

From: The dynamics of hematopoiesis over the human lifespan

Extended Data Fig. 5

(a) Stability and error values for cNMF trials using the indicated number of component gene expression programs (k) noting that 35 component programs were used in the final analysis. (b) Correlation matrix comparing Euclidean distance between the 35 component GEPs indicating the independence of the 35 programs from each other. (c) Spearman coefficient correlation matrix comparing the 35 GEPs. Note that the highest Spearman coefficient value is only 0.59, orthogonally indicating the independence of the 35 GEPs from one another. (d) Gene set enrichment analysis for the indicated lineage signatures against the identified lineage GEPs. Colored shading corresponds with nominal enrichment score (NES) and circle size with false discovery rate (FDR) q value. (e) Normalized average GEP usage per cell was mapped for each sample across the human lifespan, with LOESS fit curves indicating average utilization.

Back to article page