Figure 2.

Misfit in M. The error between portions of the M and D matrices is shown as the dimensionality increases for the Rosetta data as fit by LS-NMF. The bottom 25% of the genes in terms of precision (i.e., highest variance) are shown together with the top 25%. Since LS-NMF relies on the variance, the algorithm fits the reliable data with improving accuracy, while ignoring poor data in terms of RMSD as the dimensionality increases.