Fig. 5

scVAEDer accurately detects master regulators during cellular reprogramming. a The reverse process of DDM allows the generation of high-quality reprogramming samples from random Gaussian noise (the quality of generated samples is critical for downstream analysis). b UMAP visualization of the data latent embedding, colored based on their state (red: failed, blue: reprogrammed). c Data generated by interpolating between reprogrammed and failed states, represented by red dots. Remarkably, none of the interpolated samples are found outside the real representation of data. d Correlation values between the new interpolated samples and the average gene expression of reprogrammed and failed cells, which demonstrates the shift in gene expression between failed and reprogrammed states. e Ranking of genes based on high expression differences between t = 0 and t = 3000. f Illustration showing the process of computing gene velocities along the interpolation path to detect master regulators. g Gene set enrichment analysis using 400 genes with the highest velocities (fast responders), which reveals pathways that are crucial during the cell reprogramming process