Understanding the extent to which genomic alterations compromise the integrity of the proteome is fundamental in identifying the mechanisms that shape cancer heterogeneity. We have used quantitative mass spectrometry to characterize the proteomes and phosphoproteomes of 50 colorectal cancer cell lines that serve as a model of variable tumour clonal expansions. We have used the variation in abundance between 12k proteins and 27k phosphosites to classify the cell lines into subgroups that show distinct yet overlapping organisation with other genomic classifiers. Tightly organised protein co-variation networks deconvoluted the complexity of the molecular landscapes, revealed the proteome-wide interdependencies of biological processes and identified collateral effects of mutations on protein complexes. Proteomic analysis of human iPS cells engineered with gene knockouts confirms the transmission of co-variation as a characteristic of protein interactions and exposes that stoichiometry is tightly regulated at the protein level. Proteomics, genomics and gene expression models all appeared to be largely mutually exclusive in predicting drug responses for 265 compounds further highlighting the utility of proteomics. Overall, we provide a deep integrative view of the molecular structure underlying the variation of colorectal cancer cells.