Smart Pooling of mRNA Samples in Microarray Experiments
Raghunandan M. Kainkaryam1, Anna C. Gilbert2, and Peter J. Woolf1. (1) Chemical Engineering, University of Michigan, Ann Arbor, MI 48109-2136, (2) Mathematics, University of Michigan, Ann Arbor, MI 48109-1043
Pooling mRNA samples in microarray experiments refers to the act of mixing mRNA from different samples and testing that mix on a microarray chip. The usual motivations for pooling include reducing biological variation, saving on chips, or lack of sufficient mRNA from samples. Recent literature suggests that pooling mRNA samples may be possible but there are a number of theoretical and experimental issues to be explored. These studies employ simple pooling strategies where similar samples (replicates) are pooled to obtain an average value of gene expression. In this talk, we propose a smart pooling strategy, called poolMC, which would test a number of different samples (n) across several chips (m, where m < n) and provide robust and accurate gene expression for all samples, while saving on chips used. poolMC achieves this by testing each sample multiple times in different combinations with other samples and decodes the gene expression for all genes in each sample via linear programming. poolMC is based on state-of-the-art algorithms from the field of compressed sensing. The application of poolMC to a microarray experiment requires the satisfaction of two conditions, namely linearity and sparsity. poolMC requires that the intensity measurement from a chip with pooled mRNA, for each gene, represent a linear sum of the contributions of that gene from each sample mixed on that chip. The sparsity condition provides the savings in number of chips by requiring that a gene's expression across all samples (n) being tested is mostly unchanging. Testing portions of each sample on multiple chips ensures that the decoded gene expression is robust to measurement error. We describe the theory motivating poolMC, the algorithms for experiment design and decoding of gene expression and the implementation features specific to microarrays. Finally, we demonstrate poolMC's ability to robustly and accurately quantify gene expression values for small and large sample sizes through an in-silico experiment using real gene expression data. In summary, poolMC provides a new paradigm for measuring gene expression from several samples while providing cost savings and robustness to measurement error, subject to the constraints mentioned above.