|
| |
|
By looking at gene expression, scientists are searching for physical evidence that a gene has been
"turned on" or activated. This data is often presented in a visual format as two-dimensional images,
which can obscure the valuable information present in the original three-dimensional organism.
As part of a collaboration with academic researchers, MathEcology developed a group of custom
mathematical algorithms which extract gene expression from two-dimensional images and represent the
information in three dimensions:
|
| Drosophila Melanogaster Embryos |
| Embryonic Chick Limbs |
| Whole-Mouse Physical Sectioning |
| Embryonic Mouse Somites |
| |
|
| |
 |
Our goal was to process digital images of the egg of the Drosophila melanogaster, which is a regular
ovoid, stained for the expression of a single gene and manually oriented with the anterior-posterior
axis corresponding to the horizontal, the dorsal-ventral axis corresponding to the vertical, and
anterior to the left, dorsal to the top. The custom algorithm developed by MathEcology scientists
assigned spatial coordinates (x, y) and RGB values to the pixels of the original digital images based
on the shade and intensity of the gene expression stain within that pixel.
|
|
Signal intensity of the gene expression was calculated and represented numerically based on a detailed
system of equations such that black has the highest signal intensity and white has the lowest. A
unique threshold was set for each image, which eliminates pixels with low or no gene expression.
The spatial and signal intensity data contained in the point coverage was rescaled so that the x- and
y-coordinates correspond to percent egg length and percent egg height, respectively, and so that gene
expression is represented by integer values. These points were plotted in three dimensions to produce
three-dimensional graphs of gene expression.
|
 |
 |
As an extension of the project, we developed a generic three-dimensional model drosophila embryo by
extruding a flat image into the z-dimension based on the known geometry of the egg. Then, for each
gene expression pattern of interest, we draped the dataset determined by the process above over
the three-dimensional model to create a new set of coordinates containing x, y, z, and gene expression
values. These new models allow for viewing of the gene expression data from a variety of angles and
alignments, and can be animated to show depth.
|
| [top] |
| |
|
| |
 |
Hind limbs of embryonic chicks were physically sectioned and stained for histology, then photographed
using a transmitted light microscope. Once each image in the resulting series was corrected for noise,
alignment and contrast, we sectioned out the bone structure and created two coordinate files defining
the data contained in the image series. Utilizing custom algorithms, the pixels of the original
digital images were assigned spatial coordinates (x,y) and RGB values based on the shade and intensity
of the histology stain. A unique threshold was set for each image series, which eliminated pixels
containing no histology information.
|
|
| |
 |
|
Z-coordinates for the points within each image were calculated based upon the thickness of each physical
slice and the position of that image within the series.The two coordinate files -- one containing data
for the entire limb, the other describing only the bone structure therein -- were merged to create one
large database with multiple fields describing the limb histology. As an extension of the original
algorithm, we created a custom application to visualize the data file in three dimensions. This
application allows the limb to be viewed from a variety of angles and alignments, and to adjust the
opacity / transparency of the components of the limb in order to illustrate various aspects of the
bone structure.
|
 |
| [top] |
| |
|
| |
|
An embryonic mouse was physically sectioned and stained to show gene expression, then photographed
using a transmitted light microscope. Each image in the resulting series, though prealigned via
external markers, contained a great deal of background noise and showed an insignificant degree of
contrast in the staining. After correcting for noise and contrast, we sectioned out the gene
expression and created two coordinate files defining the data contained in the image series. Utilizing
custom algorithms, the pixels of the original digital images were assigned spatial coordinates (x,y)
and RGB values based on the presence of gene expression.
|
 |
|
|
A unique threshold was set for each image series, which eliminated pixels containing no gene expression.
Z-coordinates for the points within each image were calculated based upon the thickness of each
physical slice and the position of that image within the series. The two coordinate files -- one
containing data for the entire mouse embryo, the other describing only the gene expression therein --
were merged to create one large database with multiple fields.
|
|
Based on the original algorithms, we created a custom application to visualize the data file in three
dimensions. This application allows the embryo to be viewed from a variety of angles and alignments,
and to adjust the opacity / transparency of the components of the model in order to illustrate various
aspects of the gene expression.
|
| |
 |
| |
| [top] |
| |
|
| |
|
Somites from an embryonic mouse were fluorescent-stained for gene expression, and the optical sections
obtained using a confocal laser microscope were captured in digital images (which showed only the gene
expression, not the structure of the somite itself). The image series was processed via custom algorithms
developed by MathEcology scientists in a manner similar to that described above for the whole mouse
embryo. The three-dimensional model of the somite gene expression allows for previously unavailable
visualization of the structure and distribution of a number of different genes, and offers new insight
into the process of somitogenesis.
|
 |
| [top] |
|
| |
| |
| |