Comp/Phys/Mtsc 715 Potential Projects Spring 2008

Project #1

Client: Tom Craven (tom@unc.edu), UNC Geography.

Team: Kwangbom Choi, Thomas Craven, Yorick de Visser, Daniel Fovargue.

Instructor: Russ Taylor.

Description: Satellites provide multispectral imagery of the surface of the Earth, taken at wavelengths from radio waves through ultraviolet. Each wavelength provides a snapshot of the land surface. Together, in various combinations, the images can provide information about what lies on or just beneath the surface of the Earth. These are joined by magnetic maps of the surface and by cartographic information. The problem is how to determine what particular combinations of these many variables can be used to determine important features of the surface: how much is forest, how much grassland, and so forth.

We're looking for a visualization tool to help us figure out which of the many variables, and in what combinations, will be effective indicators. One approach that we already take is to reduce the number of signals using Principal Component Analysis methods to bring the data sets down from hundreds to tens. But we still need to understand how these covary across the surface of the Earth, together with ground-truth measurements.

Project #2

Client: Imran Shah (Shah.Imran@epamail.epa.gov), US Environmental Protection Agency.

Team: Lauren Cooper, Gregory Herschlag, Hangchin Yeh.

Instructor: Russ Taylor.

Description: I'd like to propose a visualization data set on multi-scale dynamic tissue simulation, which is part the Virtual Liver Project at the US EPA in RTP. Our objective is to simulate the dynamic biological response of the liver to chemical exposure to understand chronic injury due to environmental agents. We're focusing on the functional unit of the liver -- the hepatic lobule. The lobule structure is being represented in a hybrid 2D mesh with chemical gradients and a cellular network.

I've emailed a not-for-public-display PDF file describing the problem in more detail, with some pictures.

Data:

Project #3

Client: Dan Reichart, UNC Physics & Astronomy. Contact: Adam Trotter (atrotter@physics.unc.edu)

Description: We are modelling the time-dependent spectra of gamma-ray burst afterglows, using photometric data obtained from several instruments at a variety of frequencies (from infrared to X-ray) at various times (from a minute to several days after the initial burst). We have developed a Bayesian genetic algorithm to fit a physical model to the time-dependent photometric data from each individual burst, i.e., apparent magnitude as a function of frequency and time. We are interested in finding a visually intuitive way to compare these fitted models with one another, to identify time- and frequency-dependent signatures that might characterize different physical classes of GRBs, and the occurrence of certain interesting physical phenomena (e.g., dust destruction, molecular hydrogen dissociation) in the environment of the GRB.

Data: XXX.

Project #4

Client: Tom Clegg, UNC Physics & Astronomy/TUNL. Contact: Billy Mohr (billy.mohr@gmail.com)

Team: James Martindale, William Mohr, Jeff Poole, Adam Trotter.

Instructor: Cory Quammen.

Description: At the Triangle-area University Nuclear Laboratory (TUNL), the FEL is a Free-Electron-Laser that produces a gamma-ray beam about 2" in diameter with a flux of 10^7 per second, called the High-Intensity-Gamma Ray-Source (HIGS). I am currently working on an imaging system to characterize the intensity and profile of the HIGS beam at the FEL. Currently, it is not completely well understood what the energy resolution and intensity profile looks like across this 2" diameter. In addition, the technicians "steer" the beam based on indirect measurements and intuition instead of direct knowledge of exactly where the centroid of the beam is. While the physics aspect is designing and implementing a useful detection system that would do this, a visualization of the data analysis to demonstrate qualitatively how the beam looks and is behaving would be extremely useful. I already have some data from two different systems that I am in the process of analyzing, coding (it's slow going), and trying to understand where the physics actually begins and the detector response/physical constraints end.

Here is a picture of a preliminary visualization that I've done. It's very rough but I think it demonstrates well the beam intensity and position. This was taken behind a 1" collimator. Obviously the centroid of the intensity is not centered directly down the middle, which is where we'd like it to be. Energy resolution is a secondary concern, but not one that we're overlooking. We are more concerned with position resolution of how accurately we can steer and measure the beam, specifically how well the detector's position resolution is. The visualization project would be useful because 1)I don't have a good way of picturing a 2D scalar spectrum immediately. It took a bit just to get this one picture. One of the hopes of this project is that it would take live data and have that flashlight image within seconds/mintues.

Also, although it may not be obvious from the picture, the actual structure of the detector, (a position-senstivite segmented PMT (photo multiplier tube) split into 8x8 pixels about 1/4"x 1/4" per pixel) becomes apparent and distorts the gaussian-like shape of the beam we would expect which is evident in the 2nd diagram. The 2nd diagram is a 1D projection in the x-axis. The valleys are evenly spaced and related to the pixel structure. Although I'm not sure a visualization of a grid of the pixels superimposed upon the spectra would solve that problem, it would certainly help.

Click on any of the images for a higher-resolution version.

Project #5

Client: Mike Falvo (falvo@physics.unc.edu), UNC Physics & Astronomy.

Team: Shengqian Chen, Zachary Koterba, Huai-Ping Lee.

Instructor: Cory Quammen.

Description: Our lab is doing experiments on biological membranes, specifically cell membranes and deforming sheets of fibrin. We use an atomic-force microscope to manipulate the specimens and a fluorescent optical microscope to watch the resulting deformations. The movie to the right shows an example of us pulling on a single fiber, but the data sets of interest for this project are actually sheets.

We run feature-tracking software developed by Brian Eastwood to extract the motion of each portion of the sheet and to compute the strain tensor at every point on the surface. We now want to understand what this is telling us about the material properties of the cells and fibrin sheets.

Data: We have tracked data sets on fibrin sheets suspended over gaps like the one seen in the image above and over circular openings. We have recently acquired (and not yet tracked) cell membrane deformation when the cell is pulled on using magnetic beads. We'd also like to see what the visualization does when the tensor fields are constructed by algorithm, to test our intuition of what strain does in known cases.

Note: Russ works with this project, which provides a conflict of interest. We believe that this conflict is manageable by having Cory supervise a team that selects this project.

Project #6

Client: Yueh Lee (yzlee@physics.unc.edu), UNC Radiology, UNC Physics & Astronomy.

Description: We are developing a carbon nanotube based 4-D micro-computed tomography system for the imaging of mice. Our data consists of iso-tropic high resolution 3-D data sets at multiple time points in a respiratory or cardiac cycle. We have the unique ability to perform prospective gated imaging for both cardiac and pulmonary signals, utilizing a carbon nanotube based x-ray system.Currently, there is no other system in the world capable of gating with the same flexibility of our system.

While standard axial and reformatted allow us to visualize pathology, the subtle changes of a normal cardiac and respiratory cycle are nearly impossible to see on standard 2-d views. The goal of the software would be to improve the 4-D visualization of micro-CT for the heart and lung, and incorporate analysis software to characterize cardiac and pulmonary diseases. This software will likely be incorporated into the beta version of the system to be used in collaboration with multiple labs on UNC campus and offers the opportunity for multiple future publications.

Click on the images for higher-resolution versions of each.

Data: 3D CAT scans of mice over time.

Project #7

Client: Leonard McMillan (mcmillan@cs.unc.edu), UNC Computer Science (on behalf of collaborator).

Team: Kyle Moore, Catherine Welsh.

Instructor: Cory Quammen.

Description: Visualizing all Identity-By-Descent (IBD) genomic regions between related strains an/or subspecies. The concept is simple, each strain has its own unique DNA sequence, and we have already developed code that finds all of the shared subsequence intervals amongst all strain subsets (a mouthful). The size of this data set is hard to imagine. Each animal (mice in this case) has a sequence of ~2.5 billion bases, that differs at ~8 million positions.

I have hacked a few prototypes to explore tiny fractions of the data, but none are completely satisfying. To the right is an image from the 16- strain set showing the shared intervals between all pairs (2-strain subsets) involving a single specified strain and a single chromosome. (Click on the image for a higher-resolution version.) The color shade indicates the marker density of the interval in the other strain (a key is shown across the top), and the uniformly colored bar indicates any/every interval where there is evidence of IBD. There is also a tiny histogram (in green) atop each row showing the actual marker density. I have made similar diagrams for all strain- pairs and chromosomes, but, in terms of the data, this is only the tip of the iceberg. Trips, (subsets of 3) and other subsets are very interesting to biologists. I think that there is some potential for using 3-D in this problem, as well as alternative data depictions -- all of which could be explored in a project.

Data: I am working with 2 data sets, one with 16 strains, and a second with 70. The 16-strain set has 65534 interesting subsets (2^16 - 2), and the 70 strain set has > 10^21 subsets. Our goal is to interact with and explore all of them in some fluid way. The data also has additional variables that we would like to overlay (i.e. genes and marker density).

Note: This is more of an information-visualization application than a scientific-visualization application. This will result in a team that works on this project doing a bit of independent study to be able to select and evaluation from among potential solutions to the problem. The instructors can provide pointers to the relevant literature (some of which is linked from the course home page) to help with this.

Project #8

Client: Leonard McMillan (mcmillan@cs.unc.edu), UNC Computer Science (on behalf of collaborator).

Description: Finding evidence of historical mutations, deletions, copy-number variations, inversions, and recombination events that resulted in a given sequence. In essence this process attempts to parsimoniously run evolution backwards. The simplest model of evolution (called the infinite sites model) imposes a structure on the genome. In particular, pairs of biological markers (variations in the base sequence or single-nucleotide polymorphisms (SNPs)) should exhibit only 3 of 4 possible states. Of the forces shaping the genome only two can upset this rule -- multiple mutations at the same site (homoplasy) and recombination. Finding out where these events might have occurred is of great interest to geneticists, and it underlies our ability to assign genetic links to phenotypes (such as diseases susceptibility). The process of finding these incompatibilities is interesting in terms of both computation and visualization. It involves evaluating all pairwise relationships among markers which is a huge matrix. For example, we are working with an 8 million marker set which implies an 8M by 8M matrix (more than 10^13 cells). We'd like to be able to examine this matrix as an image, dynamically (i.e. with its elements computed on the fly). Click here for an animation that has already been made.

Data: We have techniques for computing an accurate filtered version of the matrix at any scale. Our goal in this project would be to construct a tools for real-time exploration (zooming into and out of an 8M x 8M image that is computed on the fly). I have attached a second figure to illustrate a small example of this problem. Click on the image thumbnail to get the full-sized image.

Notes: (1) This is more of an information-visualization application than a scientific-visualization application. This will result in a team that works on this project doing a bit of independent study to be able to select and evaluation from among potential solutions to the problem. The instructors can provide pointers to the relevant literature (some of which is linked from the course home page) to help with this. (2) The project goal will be modified from the stated "construct a tool for real-time zooming" to "design the best visualization technique to help answer the questions on this data set" to qualify for a course project. I'm in contact with Leonard to make sure this is acceptable.

Project #9

Client: Matthew Berginski (mbergins@unc.edu), on behalf of Eric Vitriol (Klaus Hahn's lab, Department of Pharmacology).

Team: Matthew Berginski, Michael Elder, Vinay Swaminathan.

Instructor: Russ Taylor.

Description: The machinery that helps cells move involves the interaction of a wide range of proteins. One structure that contains many proteins and is crucial for cellular motility is the focal adhesion, which acts as the anchor to the outside environment during migration. Focal adhesions consist of protein complexes (read as a glob of proteins) that transiently come together and break apart as the cell moves. One protein that is associated with all focal adhesions is Paxillin. The data set consists of time-lapse images of single cells, with the value at each pixel representing the concentration of Paxillin (see image to the right, click on it for a larger image). Image processing routines have been developed to automatically identify the adhesions in a given image and track each adhesion through time. Several properties of each adhesion have also been gathered and it would be interesting to visualize the data. Of particular biological interest will be any correlations between parameters, such how changes in area are associated with changes in Paxillin concentration or how the location of adhesion birth relates to speed. Since this data is spacial, there may also be unexpected trends in these variables that divide the cell into particular regions.

Click here to see a movie of the cell in action. Movie Key: On the left is an outline of all the identified cellular features. The outside edge is the outside edge of the entire cell, while the small circle inside the cell are the outlines of the focal adhesions. As the movie plays the outlines from all of the movies stack onto one another, with time increasing going from blue to red. On the right is the Paxillin concentration image with all of the focal adhesions outlined. Each color is assigned to a single adhesion lineage and stays with that lineage until a merge or death event, although I am still tweaking the tracking software.

Data:

Project #A

Client: One of Tom Craven's professors in Geography, UNC Geography.

Description: Ecohydrological modeling and global climate change using a spatially-explicit model he developed called RHESSys.

Data: XXX.

Project #B

Client: If we can't find enough, we'll have a team look at the IEEE Visualization Design Contest problem and pick a subset of the questions. This will be suboptimal because (1) you won't get to directly interview the scientists, that has already been done) and (2) you can't submit to the contest in any event because I'm running it this year. It is pretty clear that we're going to have enough projects, so nevermind.

Description: http://2006_ieee_vis.sdsc.edu/2006_ieee_vis_data/2008