It is clearly possible that medial remodeling can permit enolases of different bind ing preferences to be distinguished, despite significant conformational differences, like a closed active site. For tyrosine kinase structures with small find more info gatekeeper residues remodeled onto the templates 1qcf and 2hz4, the median volume of the largest fragment between the template and the model was frequently smaller than the fragment volume computed with queries from large gatekeeper amino acids. The median volume of the largest fragments between the ATP binding cavity of 1qcf and all of the modeled ATP bind ing cavities of the large gatekeeper tyrosine kinases were statistically significant, but the median volume of 11 of the modeled binding cavities from 26 tyrosine kinases were also statistically significant.
Variations in tyrosine kinase binding cavities were much larger than among enolases, and the difficulty of this classification problem is apparent from the 11 incorrect predictions here. For example, the modeled kinase 2SRC exhibits cavities ran ging from near zero to 1400 3, primarily because of the great diversity in models for 1T45 and Inhibitors,Modulators,Libraries 2SRC. Medial remodeling on tyrosine kinases based on the 2hz4 tem plate produced similar results Medial remodeling elimi nated models where the binding cavity was extremely dissimilar, and, approximately half the time, models of cavities with similar binding preferences were more similar to the template than those with different Inhibitors,Modulators,Libraries binding preferences. Patterns of statistical significance revealed a similar trend.
These results, taken at a medium scale, suggest that medial remodeling can produce Inhibitors,Modulators,Libraries effective predictions, as in the enolase superfamily, but remodeling may not be as successful for very diverse superfamilies, like the tyrosine kinases. These results also demonstrate that median remo deling Inhibitors,Modulators,Libraries can eliminate structural outliers caused by the non deterministic nature of structure prediction algorithms while reducing errors from conformational change. Conclusions We have demonstrated simple and medial remodeling approaches for comparing binding sites in flexible pro teins. While most algorithms for comparing Inhibitors,Modulators,Libraries protein structures are focused on the selleck identification of remote homologs, we seek to predict structural determinants of specificity among closely related proteins. Our approach exploits the relatedness of our datasets by using structure prediction algorithms to compensate for conformational flexibility. Since homology modeling is most accurate when predicting structures that are similar, our approach strongly complements the intended application. We demonstrated our results on sequentially nonre dundant datasets representing the enolase and the tyro sine kinase superfamilies.