 |
 |
|
|
|
|
|
|
||||||
 |
|||||||||||||
- Computational Methods for Biological Modeling and Simulation
- Mentor: Dr. Russell Schwartz
- Flux Balance Analysis of Metabolic Network of Mycobacterium Tuberculosis to identify potential drug targets
Team members: Pavitra Athakitkarnka, ChengNing Ko, Soumya Luthra
Tuberculosis remains a major global health concern and there is a need for the development of effective drugs to fight the disease. In this study, Flux Balance Analysis (FBA) approach is used to study the role of each enzyme in the context of entire metabolic network. The essentiality and potential of being anti-tuberculosis drug target for each gene product is analyzed. Since Aromatic amino acid biosynthesis has been found to be essential for M. Tuberculosis, but can’t be found in mammals, we set our objective function to the rate of Tyrosine (TYR) synthesis. We verify the usability of the model to our goal by employing Robust Analysis and Flux Variability Analysis (FVA). As a result of our study, 26 out of 1025 enzymes have been found to affect TYR Synthesis by more than 20%. Inhibition of 11 of the 26 enzymes has lethal effect on TYR synthesis which qualifies them as potential targets of anti-tuberculosis drugs. These include Shikimate Kinase (SHKK) which is a well known anti-tubercular drug target. The study demonstrates the application of FBA on metabolic network for rational identification of potential drug targets.
- Machine Learning
- Mentor: Dr. Carlos Guestrin
-
Tensor Decomposition and Classifiers to Predict the Human Mind
Neural activities in the brain, measured by functional Magnetic Resonance Imaging (fMRI), seem to support particular cognitive processes. If we can successfully classify fMRI signals, it will demonstrate that it is possible to predict associations between observed fMRI data, and cognitive processes. In this project, I trained Logistic Regression classifiers on fMRI datasets, and then compare the classification results with those obtained from Guassian Naïve Bayes classifiers. As a result of this study, LR classifiers performed better than Guassian Naïve Bayes classifiers in average ranking and accuracy. Also, GNB classifiers performed better than random by average ranking, but performed worse than random by accuracy. (I unsuccessfully tried to decompose the brain fMRI dataset into a three dimensional tensor. I only discussed about tensor decomposition in Materials and Methods section.)
- Computational Genomics
- Mentor: Dr. Eric Xing
- Experimental Techniques in Molecular Biology
- Mentor: Dr. Carrie Doonan
- Mapping Lys2 Mutants by Marker Rescue
Lab partner: Britt Weston-Ball
The significance of this project is to learn the techniques used in genetics lab. In this experiment we wanted to map the mutation location on a lys2^- mutant strain of yeast. Alternatively, we could, instead of marker rescue, sequence the chosen strain to find the point mutation responsible for lys2^- phenotype. However, this sequencing method is relatively expensive. On the other hand, marker rescue is a technique that is simple and fast, has high success rate, and does not require a lot of instruments. Our marker rescue technique maps the mutation location by transformation and plating. At the end of the experiment, we were able to locate the mutation point responsible for lys2^- phenotype. Marker rescue occurred and LYS2 gene was restored in some transformations with some plasmids. But, marker rescue didn’t occur in other transformations. Thus, the lys2^-mutation point must lie in the partial LYS2 gene carried by the former group of plasmids while NOT lie in the partial LYS2 gene carried by the latter group
- Molecular Biology
- Computational Biology
- Advanced Genetics
 |
Class Projects |  |
We have used microarray technology to investigate the gene expression profiles at transcriptional level over a period of time. In this critical review, I discuss some respects of the static unsupervised clustering algorithms for microarray data, and then discuss the reason why we need the transition from using static clustering to time-series evolving network algorithms to understand the roles of individual genes in biological processes.
|
Project Proposals | |
I compare two of the clustering algorithms, hierarchical clustering and k-means clustering, used to process microarray data. In a single experiment, microarrays can measure the expression levels of thousands of genes. After grouping these genes, we expect that the genes in a same group would potentially exhibit the common function. In order to partition these genes into groups with respect to their similar expression patterns, several clustering algorithms have been used.
|
Presentations | |
