Three-dimensional quantitative structure-activity relationships (3D-QSAR) are aimed at building quantitative models by relating biological activities of a series of ligands to their 3D properties. In general, 3D-QSAR require the determination, or an educated guess, of the bioactive conformation of a template molecule, followed by alignment of the whole dataset onto the latter. The resulting PLS model may have the power to predict the activity of new molecules before they are synthesised and tested. However, identifying the bioactive conformation of a template molecule is not a trivial task, and even when it is known, usually from an experimentally determined structure of the ligand-target complex or from carefully designed and synthesized rigid analogues, the alignment procedure itself is a difficult and time-consuming operation, especially in the presence of flexible or structurally heterogeneous ligands. When the structure or even the identity of the target is not known, it becomes difficult to hypothesize a univocal and reliable alignment, thus making 3D-QSAR hardly feasible. Unfortunately, the lack of knowledge of the structure of the target is also the situation where a ligand-based approach would be most desirable. Herein, a method for predicting the binding mode of a series of ligands is proposed. The procedure does not require structural knowledge of the binding site. Candidate alignments are automatically built and ranked according to a built-in consensus scoring function, which takes into account the quality of the alignment along with the internal predictive power of the associated 3D-QSAR model. The idea of reversing the approach to 3D-QSAR models, namely using them as a tool to select the best among many possible alignments, represents a substantial shift in paradigm compared to the traditional 3D-QSAR workflow outlined above. We have implemented this procedure through two open-source software projects: Open3DALIGN for unsupervised generation and scoring of alignments [1], and Open3DQSAR for MIF computation, PLS model building, validation and refinement through variable selection [2]. To determine the feasibility as well as the domain of applicability of our approach, we chose as test bench the eight datasets gathered from literature by Sutherland and co-workers. Results show that reliable affinity and binding mode prediction of new drug candidates may be achieved, provided that they have less than six-eight rotatable bonds. [1] Tosco, P.; Balle, T.; Shiri, F. J. Comput.-Aided Mol. Des. 2011, 25, 777-783. [2] Tosco, P.; Balle, T. J. Mol. Model. 2011, 17, 201-208.
A 3D-QSAR-driven approach to binding mode and affinity prediction
TOSCO, Paolo;
2011-01-01
Abstract
Three-dimensional quantitative structure-activity relationships (3D-QSAR) are aimed at building quantitative models by relating biological activities of a series of ligands to their 3D properties. In general, 3D-QSAR require the determination, or an educated guess, of the bioactive conformation of a template molecule, followed by alignment of the whole dataset onto the latter. The resulting PLS model may have the power to predict the activity of new molecules before they are synthesised and tested. However, identifying the bioactive conformation of a template molecule is not a trivial task, and even when it is known, usually from an experimentally determined structure of the ligand-target complex or from carefully designed and synthesized rigid analogues, the alignment procedure itself is a difficult and time-consuming operation, especially in the presence of flexible or structurally heterogeneous ligands. When the structure or even the identity of the target is not known, it becomes difficult to hypothesize a univocal and reliable alignment, thus making 3D-QSAR hardly feasible. Unfortunately, the lack of knowledge of the structure of the target is also the situation where a ligand-based approach would be most desirable. Herein, a method for predicting the binding mode of a series of ligands is proposed. The procedure does not require structural knowledge of the binding site. Candidate alignments are automatically built and ranked according to a built-in consensus scoring function, which takes into account the quality of the alignment along with the internal predictive power of the associated 3D-QSAR model. The idea of reversing the approach to 3D-QSAR models, namely using them as a tool to select the best among many possible alignments, represents a substantial shift in paradigm compared to the traditional 3D-QSAR workflow outlined above. We have implemented this procedure through two open-source software projects: Open3DALIGN for unsupervised generation and scoring of alignments [1], and Open3DQSAR for MIF computation, PLS model building, validation and refinement through variable selection [2]. To determine the feasibility as well as the domain of applicability of our approach, we chose as test bench the eight datasets gathered from literature by Sutherland and co-workers. Results show that reliable affinity and binding mode prediction of new drug candidates may be achieved, provided that they have less than six-eight rotatable bonds. [1] Tosco, P.; Balle, T.; Shiri, F. J. Comput.-Aided Mol. Des. 2011, 25, 777-783. [2] Tosco, P.; Balle, T. J. Mol. Model. 2011, 17, 201-208.
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
