Fig. 1

Projection onto CRC subtypes reduces the input gene expression data set to a 6-dimensional space. The example shown here is for the LP subtyping scheme defined by Marisa et al. [17]. Starting from the full gene expression matrix (a), transformation by Eq.(4) (reproduced at the top of figure b) generates the 6 × 209 matrix of ζ coefficients shown as a red-blue heat map (b), with the corresponding discrete subtype assignments indicated by the colored bar at the top. The subC regression model uses the ζ coefficients as covariates (Eq.(5), reproduced at the bottom of figure b)