New insights into template-based protein modeling techniques

Abstract

introduction: While the development of genomic sequencing methods has greatly improved the efficiency of collecting sequence data, experimental methods to obtain structure information have been lagging significantly. In order to elucidate protein structures, researchers have developed computational structural modeling techniques such as homology modeling and fold recognition (threading). The general consensus is that homology modeling is a superior approach with templates of high sequence similarity to the desired target (>30%), whereas threading is better suited for lower (<30%) sequence similarity templates. We compared recently improved threading algorithms with homology modeling to test the validity of this consensus. Methods: The most current versions of moDelleR and I-TasseR were used for model generation. We then used common assessment criteria (n-Dope, Q-mean and pRoCheCK) to verify the validity of the models. structure comparisons were also made using Chimera’s Cα root-mean-square deviation. results: Contrary to our prior expectations, the model determined by threading showed similar or even better assessment results in some criteria compared to the model generated from homology modeling. Furthermore, the structure analysis showed that homology modeling and threading protocols yield models with root-mean-square deviations of under 2 Å when used on protein sequences that share sequence identities of at least 30% to the experimentally determined protein template. discussion: We believe that recent improvements in threading algorithms will allow for broader applications of this methodology in large-scale modeling efforts. The fully automated steps could provide time efficacy. In contrast to popular belief in the modeling community, we have shown that threading could be a competitive means of modeling rather than a mere backup method.