|
Home
CV (pdf)
[Current Research]
Publications
Education & Experience
Awards, Service, &
Memberships
Computational
Tools
Professional
Resources
Personal
MaizeGDB
Iowa State
University |
PROTEIN SECONDARY STRUCTURE PREDICTION
BY COMBINING ARTIFICIAL INTELLIGENCE AND INFORMATION THEORY
Perfection of biological analysis equipments, revealed protein sequences, and completion of the human genome project created enormous amount of unprocessed data. New data mining approaches are crucial to derive meaningful relationships at cellular, as well as molecular levels. These powerful, yet fast algorithms will establish a theoretical framework towards drug discovery to be used against genetically originated diseases such as cancer and Alzheimer's. Since protein function is closely related to its structure, the prediction of protein 3-dimensional structure from its linear amino acid sequence is a challenging necessity in structural bioinformatics to tailor protein properties.
To tackle the problem of incorporating large evolutionary information algorithms, artificial intelligence (AI) methodologies are promising candidates with their learning capabilities. Developed by Vapnik, the Support Vector Machines (SVM) approach provides a superior algorithm in binary classification applications. SVM has been employed in many areas ranging from face identification, text classification to protein localization and microarrays. Meanwhile, the information theory originally created by Shanon finds its use in processing evolutionary information for the secondary structure prediction under GOR method.
Our research focuses on increasing the current level of secondary structure prediction rate (75%) to higher levels using SVM or a successful combination of SVM and GOR methods in order to improve current drug designs, thereby increasing the quality of human life.
DYNAMICS OF ELASTIC NETWORKS USING NORMAL MODE ANALYSIS
The connectivity of residues creates differentiated normal modes effective at various time scales. Recently developed Gaussian Network Model(GNM) and
Anisotropic Network Model (ANM) can predict experimental B-factors for globular proteins. Basically, these powerful models represents residues with their alpha carbons, assumes a spring-like force field between residues, and proves that the alpha-carbon fluctuations are mainly
determined by residue coordination numbers without any reference to their chemical content.
We apply the GNM and ANM models to reveal the dynamics of protein structural mechanisms during various cellular processes, including nucleic acid-protein interactions during enzymatic activity.
PROTEIN ACTIVE SITES DETERMINATION
BY POOLING POWERFUL PREDICTION METHODOLOGIES
Detection of protein-protein and protein-ligand interaction sites is of primary importance for rational drug design, as well as for the identification of the structural and functional details of metabolic pathways and signal transduction networks. This identification process can be rendered more precise and effective by incorporating several powerful active site discrimination methods whose results can be selectively combined.
This project is a collaborative effort of many research groups in Iowa State University, and our group is responsible for active site detection using the premise that the active sites should be conserved during evolution and must have accessible surface large enough for another biological molecule to dock.
For further information, please feel free to e-mail me.
|
