|
Document Type |
: |
Thesis |
|
Document Title |
: |
Classification of Structural Protein Domain Based on Hidden Markov Model تصنيف ميادين البروتينات البنيوية المبني على نموذج ماركوفي مخفي |
|
Subject |
: |
Faculty of Engineering |
|
Document Language |
: |
Arabic |
|
Abstract |
: |
PDZ is an acronym consolidating the main letters of three proteins — post-synaptic density protein (PSD95), Drosophila disc large tumor suppressor (Dlg1), and Zonula occludens-1 protein (zo-1) [3-5].The PDZ domain consists of a sequence of amino acids usually between 80-90 amino-acids and found in the signaling proteins of microscopic organisms, yeast, plants, viruses, and animals. They have been presented to act as key players ranging from cystic fibrosis to cancer. The classification of PDZ domain in laboratory based on the chemical characteristic is a very difficult and high cost task.
Thus our aim in this study is to find an algorithm to classify PDZ domain as Class I or II based on a given sequence of amino acid. We use the hidden markov model to classify the PDZ domain sequence as it has been used for solving many problems similar to our problem for example, forecast of protein-coding districts in genome successions, demonstrating groups of related DNA or protein taxonomies.
Hidden markov model was constructed by using interaction dataset. Our dataset consisted of 78 sequences of Class I, and 37 sequences of Class II domains. We split our dataset into training and test sets. In the training phase, we used 90% of the dataset, while the remaining 10% was considered as a testing set.
The HMM model consist of two important matrices, emission matrix and transmission matrix. The emission matrix represents the probability of occurrences of each amino acid in both classes of PDZ domain while the transition matrix represents the probability of transmission between classes. We assumed that our transmission matrix is (0.5, 0.2 and 0.1). Assuming that our initial matrix is 0.5 the HMM has been calculated for both classes based on all transmission matrix that we have assumed. The minimum classification error was (0.35). We tested the decision at each state and reported the final decision based on voting process and found that the classification rate has been improved. The final improvement has been made base on the multiplication of all state probabilities. Here, the decision of class 1 is taken when by multiplication of state probabilities of class 1 is greater than class 2 otherwise the decision will be class 2
We compared the classification results with respect to three different approaches [21-23] with our HMM method. We found that, the HMM method is computationally more effective than the other three classifiers for our problem. We predicted the classes of PDZ domain with accuracies of (83.25%). With this highly enquiring result, this study could be an important step in the automated prediction of PDZ domain classes. |
|
Supervisor |
: |
Dr. Rami Al-Hmouz |
|
Thesis Type |
: |
Master Thesis |
|
Publishing Year |
: |
1438 AH
2017 AD |
|
Co-Supervisor |
: |
Dr. Adnan Memic |
|
Added Date |
: |
Thursday, August 24, 2017 |
|
Researchers
| طارق أبوشنب | AbuShanab, Tarek | Researcher | Master | |
|