Craig Gallek

Partitioning Correlation Energy in Molecular Subsystems
The goal of this project is to find a method to calculate high-level electron structure information at the speed of a low-level calculation. The correlation energy is defined as the difference between the actual energy of a system and the energy obtained by a low level calculation (Hartree-Fock in this case). The original idea for the project was to try to define the correlation energy for a system as a function of its one electron density matrix. This turned out to be harder than expected. The revised idea, on which I have worked this summer, is to instead use the two-electron density matrix. In this way, columbic interactions will be more apparent. The first step in the project is to modify an existing electron structure program (GAMESS) to calculate the two-electron density matrix. Next, the code modifications from last summer will allow a one-electron potential to be added to the density matrix. Now, it is possible to perturb the system into what it should look like based on a high level calculation. By repeatedly forcing systems calculated with low-level theory to what they should look like with a high-level calculation, it is possible to train a neural net (or a similar algorithm) to reproduce these results for similar subsystems in other molecules.

Neil Halelamien

Lateral inhibition in an olfactory bulb model
The presentation of an odor causes a combinatorial pattern of activated glomeruli across the olfactory bulb. The particular pattern of activation produced depends on the concentrations of the various chemical components of the odor, with similar odorants often activating overlapping sets of glomeruli. The olfactory bulb as a whole is fundamental to the olfactory system's primary task of recognizing and categorizing odors; these tasks are accomplished in the face of difficulties such as widely varying odor concentration for single odors and the presence of strong background odors. The circuitry of the bulb consists of reciprocal excitatory-inhibitory dendrodendritic connections between mitral cells which receive olfactory input in the glomeruli and granule cells. This circuitry is thought to give rise to lateral inhibition, the amplitude of which decreases with distance away from the active mitral cell in a roughly Gaussian fashion. The functional transformation this lateral inhibition performs on the stimuli is still unclear, but a number of hypotheses have been presented, including contrast enhancement, sharpening of tuning curves, "decorrelation" of similar odors, generation of concentration invariance and elimination of redundant information. As lateral inhibition is common throughout the neural circuits of the brain, understanding lateral inhibition in the olfactory bulb could lead to insights regarding other parts of the brain.

The goal of the project was to determine the nature of the functional transformation taking place in the olfactory bulb by constructing a computational model which would allow for the virtually unlimited simulation of odorant exposures and facilitate an in-depth analysis of the olfactory bulb as an olfactory information processor. Physiological data from the mouse olfactory bulb was used to construct this computational model, which was composed of leaky integrate-and-fire (i.e. spiking) neurons with spikes time-distributed according to an alpha function. The behavior of the model paralleled data recorded from cells in the olfactory bulb, exhibiting oscillatory activity, spontaneous spiking, and temporally evolving patterns of activation over the mitral cells. The lateral inhibition in the model was also found to cause time-dependent decorrelation of the responses of mitral cells to similar odors over the course of an odorant exposure, a behavior observed in the zebrafish olfactory bulb. Such decorrelation may allow similar odorants to be more easily distinguished.

Richard Wang
(Mentors: Drs. Tom Mitchell and Robert F. Murphy)

Caption Interpretation from Online Biological Journal Articles
Being able to interpret fluorescence microscope images can dramatically enhance our knowledge regarding the location of proteins within cells. Based upon this concept, a knowledge base system that can interpret such images is being built at Murphy Lab in order to automate the collection, organization, and analysis of the biological data. The ultimate goal of such system is to interpret fluorescence microscope images and to find fluorescence microscope images depicting particular subcellular patterns.

However, currently the system lacks the feature of understanding the relationship between a caption and the panels of its figure from any biological online journal article. Besides analyzing the pixels of a cell image directly, by interpreting its caption, the image could be understood to some extent as well. Therefore, an automated caption interpreter that can map cell types and protein names in a given caption to panel labels of the figure of that caption was developed.

One challenging task for developing the caption interpreter is the extraction of cell type and protein names from the caption. However, the most challenging task is to be able to understand the caption and identify the mapping from the extracted names to panels in the figure.

The caption interpreter shows a high precision for identifying cell type names and a reasonable precision for identifying protein names. In addition, it maps the cell type and protein names to panel labels by intelligently assigning portions of text in a caption for each label and has achieved a satisfying precision as well. Adding this interpreter into the knowledge base system should improve the precision of finding fluorescence microscope images depicting particular subcellular patterns.

Oliver Woytyna
(Mentor: Dr. David Yaron)

Trying Trimers: Extension on Aggregates of Cyanine Dye Molecules in the Minor Groove of DNA
My work focused on the conformation and geometry of cyanine dye aggregates in the minor groove of DNA. The experimental work was done completely by computational methods. The specific aim of the project was to obtain spectral information from a simulated cyanine dye trimer projected onto a helical framework. First, all previous work and information on the project was gathered from two sources. This previous knowledge was then custom fit to my specific problem. Then, work began in coding the methods to handle the dye geometry and its projection on a helix. The results of my work lie in the direct configuration interaction calculations pulled from a wide library of quantum chemical knowledge. These numbers were then converted into graphs which tell how these dye trimers are oriented in the nanotemplate of the emulated DNA.