Using machine learning methods to predict protein contact map

Protein is the crucial biological macromolecules which is consist chains of amino acid residues joined together by peptide bonds in human body. Since protein's structure is determined by its sequence, predicting protein structure from its sequence has become one of the most fundamental issue in the bioinformatics. The residue contact map of protein contains knowledge which represents three dimensional conformation of protein. In this thesis, machine learning approaches to protein contact map prediction problem based on neural networks, logistic regression, support vector machines and other a few known machine learning methods were presented. The results are compared. These results demonstrate that the Logistic Regression, Support Vector Machine, Decorate, and Dagging algorithms give the better results as protein contact map prediction on our dataset.