Systematic Variance Correction Methods for Peptide Microarray Data

Abstract

Protein/peptide microarrays are high throughput (HT) methods with the potential of investigating tens to thousands of probes in a single experiment. However, technical variance creates an inevitable challenge for their application, hence the need for pre-processing strategies. Most methods of correcting to the technical variance have been developed based on DNA microarrays, from which this technology was adopted; however, key chip design differences limit their direct implementation. Microarray designs are flexible, which allows researchers to customise their targets and quality control strategies, hence, there is a need for design-specific pre-processing frameworks. The broad objective of this study was to evaluate sources of technical variation in peptide microarray data and compare performances of technical variance correction methods. Study design and site The study was a nested non-experimental study using peptide microarray data assayed for archived plasma samples, of children and infants admitted at Kilifi County Hospital (KCH) with suspected infections. The data was used in the development of the pre-processing framework in the R software environment. Materials and methods(s) A peptide microarray chip targeting 49 infectious diseases was used for the assay and GenePix array scanner used for the data extraction. The analysis framework will be developed using the R programming environment. Findings The standard methods; local background subtraction, log transformation, combating batch effects algorithm (ComBat), variance stabilising normalisation (VSN) and linear models, did not correct the technical variance significantly from the peptide microarray data. However, background subtraction using locally smoothed background intensities, and data scaling based on scale parameters calculated from Pooled-Adult Sera (PAS) sample fluorescence intensities achieved maximum technical variance stabilisation. Conclusion and Recommendation Technical variance stabilisation in peptide/protein microarray data is achievable. Morphological spot identification should be considered while estimating local background intensities, or spatial smoothing of the estimated intensities to reduce the background intensity estimation bias.