Book of Abstracts: Albany 2007
June 19-23 2007
Discrimination of β-barrel Membrane Proteins: Comparison between Statistical Methods and Machine Learning Algorithms
Discriminating β-barrel membrane proteins from other folding types of globular and membrane proteins is an important task both for identifying β-barrel membrane proteins from genomic sequences and for the successful prediction of their secondary and tertiary structures. We have developed statistical methods and machine learning techniques for discriminating β-barrel membrane proteins using amino acid composition, residue pair preference and motifs. The information about amino acid composition could correctly identify the β-barrel membrane proteins at an accuracy of 89% and exclude globular and ?-helical membrane proteins at the accuracy level of 80% (1). The residue pair preferences and motifs have more information than amino acid composition and these methods improved the accuracy of more than 95% in detecting β-barrel membrane proteins (2). On the other hand, we have used support vector machines and neural networks for discriminating β-barrel membrane proteins. These machine learning techniques improved the overall accuracy to 92%. The sensitivity and specificity are, respectively, 89% and 94%, which indicate that the machine learning techniques excluded the globular and α-helical membrane proteins at better accuracy than identifying the β-barrel membrane proteins (3). From the comparison of statistical methods and machine learning techniques we observed that the statistical methods could identify the β-barrel membrane proteins at high accuracy while an opposite trend is observed for machine learning techniques, which correctly excluded other folding types of globular and membrane proteins (4).
References and Footnotes
M. Michael Gromiha and
Computational Biology Res. Center,