Small Molecule Docking from Theoretical Structural Models - Computational Modeling of Biological Systems - page 87

Fig. 6 Example of workflow [ 12 ] for building homology models to be used in the ensemble
docking approach
a basis of a different template, and thus the binding site is specialized to recognize
a different subset of active ligands. As a result, there is an improvement in the
probabilities of detecting “true positives” (Fig. 6 ).
Different studies using ensemble docking with experimental structures have
obtained controversial results. Some authors [ 34 ] state that ensemble docking
clearly improves the performance of the docking process, while others [ 37 , 81 , 82 ]
complain about the increase in “false negatives” and suggest that the enrichment of
the results using ensembles is not so different when compared to a good-performing
crystal structure (although the rules to select “a priori” which is a good-performing
crystal structure are not evident). The situation when using homology models is
more evident, since in this case the use of ensembles increases very significantly
the sensitivity with respect to single models, decreasing only slightly the specificity
and leading to an overall clear improvement of the docking results [ 12 ]. Figure 7
illustrates the increase in the proportion of correctly predicted binding modes
when using ensemble docking compared to single model docking—only homology
models are being considered in the figure. In this example, single models produce
moderate binding mode predictions, being able to recover 30% of correctly docked