STATISTICAL METHODS FOR THE ASSESSMENT OF WITHIN-PLAFORM REPEATABILITY AND BETWEEN-PLATFORM AGREEMENT IN MICRORNA MICROARRAY TECHNOLOGIES

Over the last years miRNA microarray platforms have provided great insights in the biological mechanisms underlying onset and development of several diseases such as tumours and chronic pathologies. However, only a few studies have evaluated reliability and concordance of these technologies, mostly by using improper statistical methods. In this project we have studied performance of three different miRNA microarray platforms (Affymetrix, Agilent and Illumina) in terms of their within-platform repeatability by means of a of random-effects model to assess contribution to overall variability from different technical sources, and interpret the quotas of explained variance as Intraclass Correlation Coefficients. Jointly, Concordance Correlation Coefficients between technical array replicates are estimated and assessed in a non-inferiority setting to further evaluate patterns of platform repeatability. Concordance between-platforms has been studied using a modifed version of the Bland-Altman plot which makes use of a linear measurement error to build the agreement intervals. All the analysis have been performed on unfiltered and non-normalized data, yet all results have been compared to those obtained after applying standard filtering procedures and quantile and loess normalization algorithms. In this project two biological samples were considered: one tumor cell line, A498, and a pool of healthy human tissues, hREF, with three technical replicates for each platform. Our results suggest a good degree of repeatability for all the technologies considered, whereas only Agilent and Illumina show good patterns of concordance. The proposed methods have the advantage of being very flexible, and can be useful for performance assessment of other emerging genomic platforms other than microarrays, such as RNASeq technologies.