General Statistical Framework for Quantitative Proteomics by Stable Isotope Labeling

The combination of stable isotope
labeling (SIL) with mass spectrometry
(MS) allows comparison of the abundance of thousands of proteins in
complex mixtures. However, interpretation of the large data sets generated
by these techniques remains a challenge because appropriate statistical
standards are lacking. Here, we present a generally applicable model
that accurately explains the behavior of data obtained using current
SIL approaches, including 18O, iTRAQ, and SILAC labeling,
and different MS instruments. The model decomposes the total technical
variance into the spectral, peptide, and protein variance components,
and its general validity was demonstrated by confronting 48 experimental
distributions against 18 different null hypotheses. In addition to
its general applicability, the performance of the algorithm was at
least similar than that of other existing methods. The model also
provides a general framework to integrate quantitative and error information
fully, allowing a comparative analysis of the results obtained from
different SIL experiments. The model was applied to the global analysis
of protein alterations induced by low H2O2 concentrations
in yeast, demonstrating the increased statistical power that may be
achieved by rigorous data integration. Our results highlight the importance
of establishing an adequate and validated statistical framework for
the analysis of high-throughput data.