Improving 2D-DIGE protein expression analysis by two-stage linear mixed models: assesing experimental effects in a melanoma cell study

ELMER FERNÁNDEZ; MARIA ROMINA GIROTTI; JUAN ANTONIO LÓPEZ; ANDREA SABINA LLERA; OSVALDO PODHAJCER; RODOLFO CANTET; MONICA BALZARINI

BIOINFORMATICS (OXFORD, ENGLAND)

OXFORD UNIV PRESS

Año: 2008 vol. 24 p. 2706 - 2712

1367-4803

Motivation: DIGE-based protein expression analysis allows assessingthe relative expression of proteins in two biological samplesdifferently labeled (Cy5, Cy3 CyDyes). In the same gel, areference sample is also used (Cy2 CyDye) for spot matchingduring image analysis and volume normalization. The standardstatistical techniques to identify differentially expressed (DE) proteinsare the calculation of fold-changes and the comparison oftreatment means by the t-test. The analyses rarely accounts forother experimental effects such as CyDye and gel effects, whichcould be important sources of noise while detecting treatmenteffects.Results: We propose to identify DIGE DE proteins using a twostagelinear mixed model. The proposal consists of splitting theoverall model for the measured intensity into two interconnectedmodels. First, we fit a normalization model that accounts for thegeneral experimental effects such as gel and CyDye effects aswell as for the features of the associated random term distributions.Second, we fit a model that uses the residuals from the firststep to account for differences between treatments in protein-byproteinbasis. The modeling strategy was evaluated using datafrom a melanoma cell study. We found that a heteroskedasticmodel in the first stage, which also account for CyDye and geleffects, best normalized the data while allowing for an efficientestimation of the treatment effects. The Cy2 reference channelwas used as a covariate in the normalization model to avoidskewness of the residual distribution. Its inclusion improved thedetection of DE proteins in the second stage.