1 Data Mining Research Laboratory, Louisiana Tech University, Ruston, LA 71270, USA.
2 LSU Eye Center, LSU Health Sciences Center, New Orleans, LA 70112-2234.
{pradeep, sdua}@latech.edu,
Synopsis. Proteins are flexible macromolecules which allow the protein backbone to change from one specific folded conformation to another. Protein folding is frequently guided by local residue interactions that form clusters in the protein core. The interactions between residue clusters serve as potential nucleation sites in the folding process. Evidence indicates that functionally important protein flexible residue interactions are governed by the hydrophobic propensities that they possess. We hypothesize that proteins processes hydrophobic residue cores and that structural flexibility between interacting cores are vital to the function of the protein in the folded state. We propose a graph theory based data mining tool to extract and isolate protein structural features that sustain invariance in evolutionary related proteins, through the integrated analysis of five well-known hydrophobicity scales over the 3D structure of proteins. This tool is designed to successfully predict physico-chemical property-flexibility relationships that have been experimentally confirmed as functionally important. We previously obtained an average accuracy of 90% in protein classification, which we will now extend to incorporate protein flexibility.
1Data Mining Research Laboratory, Louisiana Tech University, Ruston, LA 71270, USA.
2LSU Eye Center, LSU Health Sciences Center, New Orleans, LA 70112-2234.
{pradeep, sdua}@latech.edu,{cclem1, dneum1, jhill, hthomp2}@lsuhsc.edu
Synopsis. Proteins are not rigid bodies; they are flexible and constantly change shape and form to perform their biological roles. While we are intuitively aware of their constantly changing nature, we have little understanding of how the inter-residue interactions are encoded in the protein sequence and, therefore, have little understanding as to how they drive structural flexibility. To address this knowledge gap, we propose a tool to predict and analyze those regions over the primary sequence of the protein that are of functional importance using a myriad of physico-chemical properties. We hypothesize that there is a correlated characteristic among the physico-chemical properties of a protein that can be used to predict functionally significant regions using machine-learning approaches. The proposed tool contributes to our understanding of the sequence and physico-chemical relationship and paves the way for us to identify local sequence property modulations that impact protein function without changing the protein structure.
1Data Mining Research Laboratory, Louisiana Tech University, Ruston, LA 71270, USA.
2LSU Eye Center, LSU Health Sciences Center, New Orleans, LA 70112-2234.
{hsi001, sdua}@latech.edu,
Synopsis. In this presentation, we explain the implementation and functionality of a weighted rule-based image classification tool. The software tool lets the user perform classification on regular camera images and on digital mammograms. The user can choose any image already in the database or can provide a new query image for classification. The tool uses association rules, which are given weights according to intra- and inter-class presence, to represent an image. Preliminary results using the weighted rule-based technique show that the tool can be used for image classification and content based image retrieval.
1Computer Science Program, College of Engineering and Science, Louisiana Tech University, 600 W. Arizona, Ruston, LA, USA 71270; 2 School of Medicine, Louisiana State University Health Sciences Center, New Orleans, LA; 3 CCT, 218 Johnston Hall, Louisiana State University, Baton Rouge, LA, USA 70803.
{xiandu, sdua}@latech.edu; {vinaya.amatya, vsrajeshkumar2000, gkonda2, shobhit.shakya}@gmail.com; {mpatil1, gkonda2}@tigers.lsu.edu, {gallen}@cct.lsu.edu
Synopsis. In our work, we seek to measure the critical performance of unsupervised clustering by measuring criteria such as clustering error distance, stability bound, convergence, and cluster tightness. We test these criteria on typical clustering functions and algorithms (e.g. K-means and subspace clustering) using the data sets available in our depository and from websites. We also present several theoretical limitations, which need further research and application domains.
1Data Mining Research Laboratory, Louisiana Tech University, Ruston, LA 71270, USA.
2LSU Eye Center, LSU Health Sciences Center, New Orleans, LA 70112-2234.
{aalex, sdua, pradeep}@latech.edu,
Synopsis. Information fusion is about integrating information from heterogeneous sources in order to facilitate understanding or provide knowledge that is not evident from the individual sources. There is a need for information fusion in many areas, ranging from sensor-fusion, where data from multiple sensors must be fused, to management information systems, where information from a number of business processes and sources must be integrated. Most of the current tools in the information arena are fairly focused, either on particular problems and applications or on specific techniques and methods. There is a plethora of literature explaining many variations on this theme of improving the information quality, reliability, or robustness through the use of an abundance of fusion concepts and tools. However, there is little work on theoretical frameworks and generic methodologies for the development of information fusion systems. Here, we propose a general information fusion cybertool for integrating data from multiple sources.
1Data Mining Research Laboratory, Louisiana Tech University, Ruston, LA 71270, USA.
2LSU Eye Center, LSU Health Sciences Center, New Orleans, LA 70112-2234.
{ssa017, sdua}@latech.edu
Synopsis. Most real world data is continuous in nature. High throughput data domains such as bioinformatics, finance, and biomedical sciences also generate highly continuous data. Interpretability and usability are desirable while handling the continuous data. The common technique used to enhance interpretability and usability is data partitioning in which each dimension of continuous data is partitioned. Data partitioning is defined as a process of splitting every dimension of data into intervals. The partitioning of continuous data is used as a data preprocessing technique in many data analysis tasks. The usefulness of data partitioning is realized in areas such as data analysis tasks (data discretization, clustering, classification) and distributed computing environment. In our work, we present our novel data partitioning techniques, evaluate them quantitatively against other classical data partitioning techniques and discuss their classical and novel applications.
1Data Mining Research Laboratory, Louisiana Tech University, Ruston, LA 71270, USA.
2Department of Ophthalmology, Louisiana State University Health Center, New Orleans, LA, USA.
{xiandu, sdua}@latech.edu, , {devireddy, moldovan}@me.lsu.edu
Synopsis. In this project, we aim to improve the interpretation and analysis of molecular states at different time intervals by applying dimensionality reduction techniques on atomic-level simulation trajectories. The molecule-level motion will be transformed into a low-dimensional embedding of the simulation system. Such research will aid in the understanding and investigation of pore growth in lipid bilayer membranes which are in the presence of edge-active agents (see Fig. 1). As simulation trajectory originates from interaction between atoms and molecules, this bottom-up analysis is rational in terms of both global and local system dynamics.
Quantitative Characterization of Brain Tumor Growth Using Geometric Analysis
Xian Du1, Sumeet Dua1,2 and Mark A. DeCoster2,3 1Data Mining Research Laboratory, Louisiana Tech University, Ruston, LA 71270, USA.
2LSU Eye Center, LSU Health Sciences Center, New Orleans, LA 70112-2234.
3Biomedical Engineering, Institute for Micromanufacturing, Louisiana Tech University, Ruston, LA, USA
{xiandu, sdua, decoster}@latech.edu
Synopsis. In this project, we aim to characterize brain tumor growth quantitatively by using geometric features. We have developed an active contour to capture the region of interest and create a convex hull model, which will show us important tumor features. Those features that we intend to capture are tumor shape, circularity, roughness, compactness, and tumor area. Furthermore, we employ the unique weighted association rule based classifier to classify normal and cancerous brain cells.
1Data Mining Research Laboratory, Louisiana Tech University, Ruston, LA 71270, USA.
2Biomedical Engineering, Louisiana Tech University, Ruston, LA, USA.
3Institute for Micromanufacturing, Louisiana Tech University, Ruston, LA, USA.
{xiandu, skk011, sdua ,jfang, decoster}@latech.edu
Synopsis. In this project, we aim to quantify the mixing performance of the passive micromixer and to, therefore, improve efficient micromixer analysis and design. To this end, we introduce a workflow of quantitative evaluation for micromixers. We are interested in evaluating the degree of uniform mixing integrated from both the information on the distribution of colored particles and the flow of particles. We make quantitative mix analysis of simulation results in order to understand the detailed mixing behavior of micromixers and to facilitate efficient micromixer design. We plan to implement a statistical evaluation of image intensities captured by microscopy which we believe will be useful to evaluate simulation results and design optimization of micromixers. Color particle tracking is applied to obtain a distribution of different types of particles.
