Systematic analysis of accuracy of homology models of proteins belonging to different structural classes

In the absence of experimentally determined high‐resolution structures, homology modeling continues to be the method of choice for retrieving structural information of proteins. However, models are not accurate and alignment error has been shown to be the most important factor affecting the accuracy of homology models. In our continued effort of large‐scale characterization of homology model accuracy, we note that model accuracy of protein belonging to different structural classes follows the trend all‐α > αβ > all‐β suggesting that backbone geometry may also influence accuracy, as the prime difference between structural classes is the backbone geometry. Surprisingly, the geometric‐factor directly contributes negligibly to the observed results. Alignment accuracy instead contributes maximally to the observed model accuracy and follows the same trend all‐α > αβ > all‐β suggesting a relationship between backbone geometry and alignment quality. Differences in the information content of sequences between the structural classes lead to the differences in the alignment accuracy enabling a priori accuracy estimates of models in a class‐dependent manner.