3D PROTEIN STRUCTURE PREDICTION USING PROBABILISTIC METHODS

Abstract

Proteins are the basic essentials of living matter and provide many biological functions that are in a wide scope ranging from biological catalysts, immunoglobulin, transfer proteins, regulatory proteins, structural proteins, movement proteins to being nutrient proteins. Each function of a protein will depend on how a protein folds and protein structure prediction is the route to understand how the protein is folded through the secondary structure. Every secondary structure formed by the amino acids is heavily dependent on the biochemistry theory of which hydrophobic interactions will play a part on whether α-helix, β-sheets or coil will be formed. In this study, a probabilistic mathematical model was developed based on Bayesian theorem. Together with a window size of 3 and using empirical data of conformational preferences of the formation of β-sheets, the method was trained on 20 related protein sequences. Using two parameters, similarity signifies the sensitivity of the method while accuracy signifies the measurement of the results to the true value. These rationales resulted in an average similarity of 65.169% and average accuracy of 67.323%. The results were further tested on non-related protein sequences and achieved a similarity of 65.445% and accuracy of 63.876%. This study presented a way of presenting a protein structure prediction method based on an empirical data on conformational preferences for protein secondary structure.