Gene Expression Deconvolution using Single-cells. J. Lindsay1, I. Mandoiu1, C. Nelson2 1) Department of Computer Science and Engineering, University of Connecticut. Storrs CT; 2) Department of Molecular and Cell Biology. University of Connecticut. Storrs CT.

   Obtaining whole-transcriptome expression profiles of closely related cell types is challenging for stem-cell biologists. Here we present an approach that utilizes single-cell qPCR probing of a small number of genes to aid in the deconvolution of whole-transcriptome profiles of mixed samples. Typically the expression profiles of a given mixture of cells is modelled as linear combination of the signature of its constituent cells multiplied by the concentration of each each cell type in the mixture. Existing approaches to deconvolution methods attempt to estimate both the cell type signatures and concentrations simultaneously, or separately if knowledge of one is know beforehand. Our method first obtains a reduced profile of constituent cell-types from single-cell samples by using k-means clustering and then averaging all cell-types in each cluster. Then we apply a robust quadratic programming method to inferring mixture proportions of mixed sample. Finally we have implemented a second quadratic program for inferring cell-type specific expression levels of genes not measured directly in single-cells based on mixture proportions derived for each mixed sample. Using real single-cell data obtained from the posterior Node-Streak-Border region of a mouse embryo and 100 simulated mixtures, a leave-one-gene-out experiment found our method estimates of concentrations had a RMSE of 0.03 and the missing gene estimates had a correlation of 0.997.

You may contact the first author (during and after the meeting) at