Linking PointCloudXplore and Matlab: Making Advanced Analysis
Easily Accessible
Problem Statement and Goals
Three-dimensional gene expression PointCloud data, generated by the
Berkeley
Drosophila Transcription Network Project (BDTNP), provides
quantitative information about the spatial and temporal expression of
genes in early
Drosophila embryos at cellular resolution. The
BDTNP team visualizes and analyzes PointCloud data using the software
application
PointCloudXplore (PCX) [1], which we developed in
close
collaboration with the BDTNP team.
The goal of this work has been to maximize the impact of BDTNP
PointCloud data by i) making PointCloud data easy to comprehend for
developers of analysis functions, ii) enabling developers to deploy
their analyses within the context of an advanced easy-to-use
visualizations system, and by iii) making novel, advanced analyses
capabilities easily accessible to biologist users.
Implementation and Results
Our approach to addressing the aforementioned challenges is to link
PCX and Matlab
1 via a dedicated interface,
thereby providing biologists seamless access to advanced data analysis
functions and giving bioinformatics researchers the opportunity to
integrate their analysis directly into the visualization application.
Figure 1(a) provides an overview of PCX and its interface to Matlab.
From a biologist’s perspective, the interface to Matlab hides the
complexity of performing analyses implemented in Matlab. The interface
allows analyses to be initiated directly from within PCX. No Matlab
knowledge is necessary to use these functions. From a bioinformatics
researcher’s perspective, the interface hides the internal PCX
architecture and requires minimal effort to make a Matlab function
accessible to PCX (see Figure 1(b)). The PCX-Matlab interface provides
a means to initiate Matlab function calls from PCX’s graphical user
interface (GUI), while automatically handling all necessary
inter-system communication including: i) export of user-defined gene
expressions, cell selections, cell locations, cell neighbors, and
additional user-definable function parameters from PCX to Matlab and
ii) import of derived gene expression channels and/or cell selections
from Matlab to PCX.
|
|
(a) Design
of PointCloudXplore. |
(b) Example
PointCloudXplore Matlab script. |
Figure
1: (a) Overview of PointCloudXplore and the interface to Matlab. (b)
M-file (top left) and PCXM header file (bottom left) of an example
function for computing the the cell-by-cell difference between two
expression patterns. Color of text is used to illustrate which parts
are mandatory (red), optional (blue), or provided in a template file
(black). The example shown on the right illustrates the use of the
function to compute the difference between late- and early-stage hunchback
expression. |
To demonstrate the usefulness of this approach, we computationally
model parts of the expression pattern of the gene even skipped.
Expression regulatory models often depend on extensive system-wide
knowledge based on years of experimental work on mutants and transgenic
constructs and specialized sets of equations and programs. With an
increasing number of components (i.e., genes), the number of potential
interactions that need to be analyzed experimentally increases
exponentially. Thus, computational methods are needed to identify
probable candidate genes for experimental verification. To address this
challenge, we implemented a genetic algorithm for finding potential
genetic regulatory interactions via optimization of a linear network
model. We implemented the optimization algorithm in Matlab and
integrated it with PCX via our cross-system interface. Integration of
the modeling with the visualization makes the modeling easily
accessible and enables biologists to define the necessary inputs
quicker and to more accurately and effectively validate the inputs and
outputs of the analysis. Figures 2 and 3 summarize the results of the
computational modeling experiment of eve using PCX and Matlab.
This work is presented in more detailed in a 2012 book chapter [2].
More details about PointCloudXplore can be found in the following
journal and conference papers [3–7]
Impact
To fully exploit the collaborative research potential of teams of
biologists, computational biologists, and computer scientists, it is
essential to overcome true and perceived obstacles for collaboration.
Biologists rarely do computation and computer scientists rarely do
biology. To maximize the impact of novel, complex, high-dimensional
data sets acquired via modern imaging or computational methods, such as
the BDTNP 3D gene expression atlas data, the data needs to be
accessible to biologists and comprehensible to developers of analysis
and visualization software.
In this work we have addressed these challenges by linking the
visualization system PCX and Matlab via a dedicated interface,
providing biologists seamless access to advanced data analysis
functions and enabling bioinformatics researchers to integrate their
analysis directly into the visualization. By being able to test new
analysis functions during development, biologists are able to provide
feedback early, facilitating communication between the developer and
the user. By utilizing PCX and Matlab, a developer can develop new
functions more efficiently without having to know anything about the
PointCloud data format or the architecture of PCX.
In our computational modeling experiments of the expression pattern of
eve,
we observed that even when using a simple linear model, the modeling
was able to predict a large range of regulators correctly for
eve
stripe 2 (Figure 2(b)), which shows that modeling can provide
interesting insights into, or at least hints at, possible regulatory interactions.
Missing inputs (regulators), noise, and limitations of the employed
computational model, however, directly affect the quality of the
predicted model and may also lead to false negatives (missing
regulators) and false positives (misidentified regulators). Modeling
results should, therefore, always be validated experimentally.
|
|
(a) Design
of the genetic network modeling experiment. |
(b) Modeling
results for stripe 2 of of the expression pattern of the gene eve. |
Figure
2: (a) Overview of the design of a network modeling experiment using
PCX and Matlab showing: (i) the set of input regulators eregs (green box), (ii) the target etarget (red box), (iii) an optional mask
specifying the area of interest (blue box), and (iv) additional input
parameters, such as the probability for crossings, pcross . (b) Overview of the modeling results for eve
stripe 2 showing: a) curve plot showing the models from 20
repeats of the experiment. All models achieved a correlation of more
than 97.x%, b) scatter plot of the target pattern and the best
model pattern, and c,d) visualization of the target and model pattern
with color indicating the relative expression (blue = low and red =
high expression). We can see that the model fits the target well and
even reproduces variations of the target stripe along the
Dorsal/Ventral (D/V) axis. |
|
Figure 3: Overview of the
modeling results for eve stripes 1, 3, 4, 5, and 6. The
correlation (corr) between the model and the corresponding
target pattern are indicated at the top-left of each plot. |
References
[1]
PointCloudXplore. http://bdtnp.lbl.gov/Fly-Net/bioimaging.jsp?w=pcx.
[2] O. Rübel, S.V.E. Ker¨anen, M.D. Biggin, D.W. Knowles, G.H. Weber,
H. Hagen, B. Hamann, and E.W. Bethel.
Linking Advanced
Visualization and MATLAB for the Analysis of 3D Gene Expression Data,
pages 267–285. Mathematics and Visualization. Springer Verlag,
Heidelberg, Germany, Jan 2012. LBNL-4891E, Presented by Oliver Ru¨bel
at the 2nd International Workshop on Visualization in Medicine and Life
Sciences 2009, Bremerhafen, Germany.
[3] O. Rübel, G. H. Weber, S .V .E. Keraenen, C. C. Fowlkes, C. L.
Luengo Hendriks, L. Simirenko, N. Y. Shah, M. B. Eisen, M. D. Biggin,
H. Hagen, J. D. Sudar, J. Malik, D.W. Knowles, and B. Hamann.
Pointcloudxplore: Visual analysis of 3d gene expression data using
physical views and parallel coordinates. In B. Sousa Santos, T. Ertl,
and K.I. Joy, editors,
Data Visualization 2006 (Proceedings of
EuroVis 2006), pages 203–210, Aire-la-Ville, Switzerland, May 2006.
Eurographics Association.
[4] Oliver Rübel, Gunther Weber, Soile V. E. Keraenen, Charless C.
Fowlkes, Cris L. Luengo Hendriks, Lisa Simirenko, Nameeta Y. Shah,
Michael B. Eisen, Mark D. Biggin, Hans Hagen, Damir Sudar, Jitendra
Malik, David W. Knowles., and Bernd Hamann.
PointCloudXplore: A
Visualization Tool for 3D Gene Expression Data, volume S-4 of
GI
Lecture Notes in Informatics, pages 107–117. Gesellschaft fuer
Informatik (GI), Bonn, Germany, June 2006. LBNL-62336.
[5] Oliver Rübel, Gunther H. Weber, Min-Yu Huang, E. Wes Bethel, Mark
D. Biggin, Charless C.
Fowlkes, C. Luengo Hendriks, Soile. V. E. Keraenen, Michael B. Eisen,
David W. Knowles, Jitendra Malik, Hans Hagen, and Bernd Hamann.
Integrating data clustering and visualization for the analysis of 3d
gene expression data.
IEEE Transactions on Computational Biology
and Bioinformatics, 7(1):64–79, March 2010. LBNL-382E.
[6] Oliver Rübel, Gunther H. Weber, Min-Yu Huang, E. Wes Bethel, Soile
V. E. Keraenen, Charless C. Fowlkes, Cris L. Luengo Hendriks, Angela H.
DePace, Lisa Simirenko, Michael B. Eisen, Mark D. Biggin, Hans Hagen,
Jitendra Malik, David W. Knowles, and Bernd Hamann.
PointCloudXplore 2: Visual Exploration of 3D Gene Expression. GI Lecture
Notes in Informatics. Gesellschaft fuer Informatik (GI), 2008.
LBNL-249E.
[7] Gunther H. Weber, Oliver Rübel, Min-Yu Huang, Anagla H. DePace,
Charless C. Fowlkes, Soile V.E. Keraenen, Cristian L. Luengo Hendriks,
Hans Hagen, David W. Knowles, Jitendra Malik, Mark D. Biggin, and Bernd
Hamann. Visual exploration of three-dimensional gene expression using
physical views and linked abstract views.
IEEE Transactions on
Computational Biology and Bioinformatics, 6(2):296–309, April-June
2009. doi:10.1109/TCBB.2007.70249. LBNL-63776.
Contact
Oliver Rübel
1 MATLAB is a registered
trademark of The MathWork Inc., 3 Apple Hill Drive Natick, MA
01760-2098, USA. Online at: http://www.mathworks.com/ .