Figure 1.

Schematic depicting the decomposition of the matrix $\mathbb{X}$ (cell spectra) into lower rank non negative matrices $\mathbb{W}$ (scores), $\mathbb{A}$ (auxiliary) and $\mathbb{S}$ (chemical bases). This method is semi-supervised when compared with traditional NMF as matrix $\mathbb{H}$ is further decomposed into two matrices $\mathbb{A}$ and $\mathbb{S}$, where A is an auxiliary matrix used to scale the data such that the mean score on each factor equates to 1 and $\mathbb{S}$ is a matrix containing Raman spectra of known biochemicals which can be fully or partially pre-specified.