Variance component analysis to assess protein quantification in biomarker discovery. Application to MALDI‐TOF mass spectrometry. Issue 2 (12th December 2017)

• Record Type: Journal Article
• Authors: Mercier, Catherine; Klich, Amna; Truntzer, Caroline; Picaud, Vincent; Giovannelli, Jean‐François; Ducoroy, Patrick; Grangeat, Pierre; Maucort‐Boulch, Delphine; Roy, Pascal
• Is Part Of: Biometrical journal. Volume 60:Issue 2(2018:Mar.)
• Page Start: 262
• Page End: 274
• Publication Date: 2017-12-12
• Subjects: biomarker discovery -- experimental design -- sum of squares type -- technological variability -- variance components; Biometry -- Periodicals; Medical statistics -- Periodicals; 570.15195
• DOI: 10.1002/bimj.201600198
• Languages: English
• ISSNs: 0323-3847

Abstract

Abstract: Controlling the technological variability on an analytical chain is critical for biomarker discovery. The sources of technological variability should be modeled, which calls for specific experimental design, signal processing, and statistical analysis. Furthermore, with unbalanced data, the various components of variability cannot be estimated with the sequential or adjusted sums of squares of usual software programs. We propose a novel approach to variance component analysis with application to the matrix‐assisted laser desorption/ionization time‐of‐flight (MALDI‐TOF) technology and use this approach for protein quantification by a classical signal processing algorithm and two more recent ones (BHI‐PRO 1 and 2). Given the high technological variability, the quantification failed to restitute the known quantities of five out of nine proteins present in a controlled solution. There was a linear relationship between protein quantities and peak intensities for four out of nine peaks with all algorithms. The biological component of the variance was higher with BHI‐PRO than with the classical algorithm (80–95% with BHI‐PRO 1, 79–95% with BHI‐PRO 2 vs. 56–90%); thus, BHI‐PRO were more efficient in protein quantification. The technological component of the variance was higher with the classical algorithm than with BHI‐PRO (6–25% vs. 2.5–9.6% with BHI‐PRO 1 and 3.5–11.9% with BHI‐PRO 2). The chemical component was also higher with the classical algorithm (3.6–18.7% vs. < 3.5%). Thus, BHI‐PRO were better in removing noise from signal when the expected peaks are detected. Overall, either BHI‐PRO algorithm may reduce the technological variance from 25 to 10% and thus improve protein quantification and biomarker validation.