conference on systems biology of mammalian cells , 2012-07-09

Title : ( Prediction of protein-protein interactions in red blood cell tissue based on protein primary structure )

Authors: afsaneh maali , Mahmood Akhavan Mahdavi ,
File:

Access to full-text not allowed by authors

Citation: BibTeX | EndNote

Question: Protein-protein interaction information in human tissue is key to understanding disease mechanisms and consequently designing of new drugs. Completion of human genome project opened up new insights into the interactions among human proteins in different tissues. However, because of the complexity of human genome different tissues have been studied separately in terms of identification of protein-protein interactions. This study aims at the computational prediction of interactions among red blood cell proteins as a basic knowledge in identification of related diseases such as leukemia. The prediction was based on the amino acid contents of proteins, the feature that has not been used in previous studies. Methods: Confident information on protein-protein interactions of human red blood cells was obtained from databases such as HPRD and PID. A collection of 23000 pair-wise interactions was prepared. Two positive and negative data sets were defined so that positive dataset comprised the interaction retrieved from databases and negative dataset consisted of equal number of interactions randomly selected from all possible interactions among red blood cell proteins. The two datasets were equally divided into two training and test sets. Using the training set a support vector machine (SVM) classifier was trained using features based upon the amino acid content of red blood cell proteins. The training was based on the notion that functionally related proteins are similar in terms of amino acid contents. For each pair of proteins, forty features pertaining to the amino acid contents of proteins were introduced. Results: The trained classifier was applied to the test set to classify protein pairs to positive and negative interactions. The average accuracy of 78% was obtained for 100 trials of random selection of negative dataset. This accuracy is comparable with other supervised methods for predicting protein-protein interactions with more number of features. Basically, the more the number of features, the higher the accuracy of prediction will be. Conclusions: A new feature was implemented to train a SVM classifier for the prediction of positive protein-protein interactions among red blood cell proteins. Although the number of features were less in comparison with more supervised methods, the accuracy of prediction was better or in the same range. This showed that the selection of an appropriate feature to train classifiers is a key factor in the identification true protein-protein interactions in computational approaches.

Keywords

, protein-protein interactions, protein primary
برای دانلود از شناسه و رمز عبور پرتال پویا استفاده کنید.

@inproceedings{paperid:1032266,
author = {Maali, Afsaneh and Akhavan Mahdavi, Mahmood},
title = {Prediction of protein-protein interactions in red blood cell tissue based on protein primary structure},
booktitle = {conference on systems biology of mammalian cells},
year = {2012},
location = {لایپزیگ, GERMANY},
keywords = {protein-protein interactions; protein primary structurs},
}

[Download]

%0 Conference Proceedings
%T Prediction of protein-protein interactions in red blood cell tissue based on protein primary structure
%A Maali, Afsaneh
%A Akhavan Mahdavi, Mahmood
%J conference on systems biology of mammalian cells
%D 2012

[Download]