Constructing 5D developing gene expression patterns without live animal imaging
- First Online:
- Cite this article as:
- Peng, H. & Myers, E.W. Biomed. Eng. Lett. (2014) 4: 338. doi:10.1007/s13534-014-0167-6
- 70 Downloads
There are five intrinsic dimensions for spatiotemporally developing patterns of gene expression, i.e. three spatial dimensions X, Y, and Z, the time, and the co-localized developing expression of multiple genes. Observing the formation of these patterns shed new light in understanding basic cellular processes and the genetic regulatory/signaling network. Ideally one would like to image this five-dimensional process in vivo, but most of current live animal imaging studies limit one to narrow time windows or small volumes or a small number of co-stained genes of interest.
Here we demonstrate reconstructing this developing pattern computationally without live imaging. For Drosophila embryos with labeled mRNA gene expression, we have reconstructed developmental time series of co-localized gene expression patterns by automatically sorting three-dimensional in situ images of late blastoderm Drosophila embryos sampled randomly from the desired time interval.
Specifically, we have developed a computational method to reconstruct such a developmental time series of the expression of a gene using 3D in situ images of a large number of Drosophila embryos sampled randomly from the desired time interval. Each sampled embryo in a data series has its nuclei labeled and two or more selected mRNA targets labeled via hybridization with probes of a different color. The multi-color images in such a series are automatically sorted into their temporal order by our new computational approach. We formulate this problem as that of learning a manifold, and solve it by first registering or aligning the images and then sorting them temporally by minimizing the alignment differences between adjacent images in a putative order. We present two approaches for ordering the data, the first based on minimum spanning trees and the second based on finding a principal curve through the data.
We have applied this computational approach to reconstruct the developmental time series of the expression of several genes in late blastoderm fly embryos.