A datamining approach to cell population deconvolution from gene expressions using particle filters
Proceedings of the 5th international workshop on Bioinformatics
Computational methods for estimation of cell cycle phase distributions of yeast cells
EURASIP Journal on Bioinformatics and Systems Biology
| Hi-index | 3.84 |
Motivation: In the study of many systems, cells are first synchronized so that a large population of cells exhibit similar behavior. While synchronization can usually be achieved for a short duration, after a while cells begin to lose their synchronization. Synchronization loss is a continuous process and so the observed value in a population of cells for a gene at time t is actually a convolution of its values in an interval around t. Deconvolving the observed values from a mixed population will allow us to obtain better models for these systems and to accurately detect the genes that participate in these systems. Results: We present an algorithm which combines budding index and gene expression data to deconvolve expression profiles. Using the budding index data we first fit a synchronization loss model for the cell cycle system. Our deconvolution algorithm uses this loss model and can also use information from co-expressed genes, making it more robust against noise and missing values. Using expression and budding data for yeast we show that our algorithm is able to reconstruct a more accurate representation when compared with the observed values. In addition, using the deconvolved profiles we are able to correctly identify 15% more cycling genes when compared to a set identified using the observed values. Availability: Matlab implementation can be downloaded from the supporting website http://www.cs.cmu.edu/~zivbj/decon/decon.html