Phenotyping Strategies

Micromolecules: definitions & links Cellular domains: Genome, preoteome & metabolome
Metabolic diversity: Scale, dynamics, & phyletics Phenotyping strategies: proteomics vs metabolomics
CMP Platforms: Platforms and workflow overview CMP Probes: The probe library
CMP Substrates: Molecular trapping & detection CMP Datasets: Data arrays for multichannel imaging
CMP Analysis: pattern recognition theory and tools CMP Exploration: N-space visualization tools
CMP Annotation: browsing & annotating data


Univariate Proteomic vs Multivariate Metabolomic Phenotyping Strategies


Univariate Proteomic Profiling Multivariate Metabolomic Profiling
Target: Proteins (metabolic, structural, etc) Target: Micromolecules
Platform: IgGs and superposition optical detection Platform: IgGs and surface optical detection
Methods: Univariate and qualitative Methods: Multivariate and quantitative
Strengths of Proteomic Libraries Defects of Metabolomic Libraries
  • large probe libraries exist
  • class resolution may be high with a single probe
  • cell shape and patterning often easily visualized
  • lower risk of nonstationarity
  • the number of probes is currently limited
  • class resolution may be low with a single probe
  • cell shape and patterning must often be computed
  • higher risk of nonstationarity
Defects of Proteomic Libraries Strengths of Metabolomic Libraries
  • libraries are species-specific, i.e. not general
  • signals are low strength & qualitative
  • classification is post hoc & gapped
  • merging classification maps is usually impossible
  • generalization across models is difficult
  • libraries are explicitly species-independent
  • signals are high strength & quantitative
  • classification is robust and spatially complete
  • merging classification maps is simple and direct
  • generalization across models is simple and direct


There are two antipodal designs for classifying cells: Univariate and multivariate. The univariate design is the standard immunocytochemical approach and requires one proven probe for every suspected class (for retina this would require > 60 specific antibodies selected from thousands of candidates) and hundreds of samples to test them all. Of course the problem is to discover the classes, and the problem of gapping quickly becomes insurmountable. The problems of a univariate strategy are many: most essential probes do not exist; data fusion is impossible; no proof of completeness or correctness is possible, a priori.

CMP is the multivariate model (the correct one for a general classification task) in which a few probes targeting overlapping classes create an N-space matrix for a single sample. The strengths of CMP are: probes for concurrent use exist; data fusion methods are robust; and mathematic completeness is possible. One need not know how many classes exist, a priori, to discover them.