| The use of high-throughput microarrays has increased significantly over the past few years in the biomedical sciences. The arrays have become denser and their applications wider. In their early versions, microarrays targeted a few thousands genomic units and, today, they can interrogate millions of them. With such density growth, analysts have been struggling with computational tools and statistical methodologies that address their scientific questions in a proper manner. Leading this surge in microarray density is the use of SNP arrays, providing the ability of obtaining genotype calls genomewide, which are later used in association studies. Motivated by this scenario, we develop a statistical methodology, CRLMM, for accurate genotyping that outperforms standard tools commercially available. This strategy is then improved with further use of hierarchical models to account for variability previously ignored by the method. We also introduce quality metrics to help the researcher assessing batch effects and SNP quality. These procedures and others were subsequently implemented in an open-source R/BioConductor package, oligo, that aims to centralize all the feature-level analyses, i.e., preprocessing methods, for oligonucleotide microarrays. Improvements on the CRLMM algorithm were implemented on the lightweight crlmm R/BioConductor package. |
