Xenbase Image ID: 159591

S6 Fig. Nonnegative matrix factorization (NMF) convergence. Convergence of NMF to a single minimum was assessed by clustering rows of the mixing matrix obtained by running NMF deconvolution with random initialization 20 times and checking that the number of tight clusters obtained equal NMF rank. This procedure shows that a single minimum is obtained up to rank 5 both at stage 12.5 (A) and stage 14 (B), although for stage 12.5, rank 4 does not lead to tight clusters. Higher ranks do not lead to a single solution, as the number of clusters recovered exceeds the rank and/or clusters are not tight (rank 6 at stage 12.5). Mixing matrices for each solution from 2 to 7 ranks are shown. Stars ( ) point out clusters that are not tight. Image published in: Plouhinec JL et al. (2017) © 2017 Plouhinec et al. This image is reproduced with permission of the journal and the copyright holder. This is an open-access article distributed under the terms of the Creative Commons Attribution license Larger Image Printer Friendly View

Return to previous page