Kinase-kernel models: accurate in silico screening of 4 million compounds across the entire human kinome

Reliable in silico prediction methods promise many advantages over experimental high-throughput screening (HTS): vastly lower time and cost, affinity magnitude estimates, no requirement for a physical sample, and a knowledge-driven exploration of chemical space. For the specific case of kinases, given several hundred experimental IC(50) training measurements, the empirically parametrized profile-quantitative structure-activity relationship (profile-QSAR) and surrogate AutoShim methods developed at Novartis can predict IC(50) with a reliability approaching experimental HTS. However, in the absence of training data, prediction is much harder. The most common a priori prediction method is docking, which suffers from many limitations: It requires a protein structure, is slow, and cannot predict affinity. (1) Highly accurate profile-QSAR (2) models have now been built for roughly 100 kinases covering most of the kinome. Analyzing correlations among neighboring kinases shows that near neighbors share a high degree of SAR similarity. The novel chemogenomic kinase-kernel method reported here predicts activity for new kinases as a weighted average of predicted activities from profile-QSAR models for nearby neighbor kinases. Three different factors for weighting the neighbors were evaluated: binding site sequence identity to the kinase neighbors, similarity of the training set for each neighbor model to the compound being predicted, and accuracy of each neighbor model. Binding site sequence identity was by far most important, followed by chemical similarity. Model quality had almost no relevance. The median R(2) = 0.55 for kinase-kernel interpolations on 25% of the data of each set held out from method optimization for 51 kinase assays, approached the accuracy of median R(2) = 0.61 for the trained profile-QSAR predictions on the same held out 25% data of each set, far faster and far more accurate than docking. Validation on the full data sets from 18 additional kinase assays not part of method optimization studies also showed strong performance with median R(2) = 0.48. Genetic algorithm optimization of the binding site residues used to compute binding site sequence identity identified 16 privileged residues from a larger set of 46. These 16 are consistent with the kinase selectivity literature and structural biology, further supporting the scientific validity of the approach. A priori kinase-kernel predictions for 4 million compounds were interpolated from 51 existing profile-QSAR models for the remaining >400 novel kinases, totaling 2 billion activity predictions covering the entire kinome. The method has been successfully applied in two therapeutic projects to generate predictions and select compounds for activity testing.     

Computational prediction of structure-activity relationships for the binding of aminocyclitols to β-glucocerebrosidase

Glucocerebrosidase (GCase, acid β-Glucosidase) hydrolyzes the sphingolipid glucosylceramide into glucose and ceramide. Mutations in this enzyme lead to a lipid metabolism disorder known as Gaucher disease. The design of competitive inhibitors of GCase is a promising field of research for the design of pharmacological chaperones as new therapeutic agents. Using a series of recently reported molecules with experimental binding affinities for GCase in the nanomolar to micromolar range, we here report an extensive theoretical analysis of their binding mode. On the basis of molecular docking, molecular dynamics, and binding free energy calculations using the linear interaction energy method (LIE), we provide details on the molecular interactions supporting ligand binding in the different families of compounds. The applicability of other computational approaches, such as the COMBINE methodology, is also investigated. The results show the robustness of the standard parametrization of the LIE method, which reproduces the experimental affinities with a mean unsigned error of 0.7 kcal/mol. Several structure-activity relationships are established using the computational models here provided, including the identification of hot spot residues in the binding site. The models derived are envisaged as important tools in ligand-design programs for GCase inhibitors.     

Challenging the gold standard for 3D-QSAR: template CoMFA versus X-ray alignment

X-ray-based alignments of bioactive compounds are commonly used to correlate structural changes with changes in potencies, ultimately leading to three-dimensional quantitative structure–activity relationships such as CoMFA or CoMSIA models that can provide further guidance for the design of new compounds. We have analyzed data sets where the alignment of the compounds is entirely based on experimentally derived ligand poses from X-ray-crystallography. We developed CoMFA and CoMSIA models from these X-ray-determined receptor-bound conformations and compared the results with models generated from ligand-centric Template CoMFA, finding that the fluctuations in the positions and conformations of compounds dominate X-ray-based alignments can yield poorer predictions than those from the self-consistent template CoMFA alignments. Also, when there exist multiple different binding modes, structural interpretation in terms of binding site constraints can often be simpler with template-based alignments than with X-ray-based alignments.     

Testing physical models of passive membrane permeation

The biophysical basis of passive membrane permeability is well-understood, but most methods for predicting membrane permeability in the context of drug design are based on statistical relationships that indirectly capture the key physical aspects. Here, we investigate molecular mechanics-based models of passive membrane permeability and evaluate their performance against different types of experimental data, including parallel artificial membrane permeability assays (PAMPA), cell-based assays, in vivo measurements, and other in silico predictions. The experimental data sets we use in these tests are diverse, including peptidomimetics, congeneric series, and diverse FDA approved drugs. The physical models are not specifically trained for any of these data sets; rather, input parameters are based on standard molecular mechanics force fields, such as partial charges, and an implicit solvent model. A systematic approach is taken to analyze the contribution from each component in the physics-based permeability model. A primary factor in determining rates of passive membrane permeation is the conformation-dependent free energy of desolvating the molecule, and this measure alone provides good agreement with experimental permeability measurements in many cases. Other factors that improve agreement with experimental data include deionization and estimates of entropy losses of the ligand and the membrane, which lead to size-dependence of the permeation rate.     

Strategies for assessing proton linkage to bimolecular interactions by global analysis of isothermal titration calorimetry data

Isothermal titration calorimetry (ITC) is a traditional and powerful method for studying the linkage of ligand binding to proton uptake or release. The theoretical framework has been developed for more than two decades and numerous applications have appeared. In the current work, we explored strategic aspects of experimental design. To this end, we simulated families of ITC data sets that embed different strategies with regard to the number of experiments, range of experimental pH, buffer ionization enthalpy, and temperature. We then re-analyzed the families of data sets in the context of global analysis, employing a proton linkage binding model implemented in the global data analysis platform SEDPHAT, and examined the information content of all data sets by a detailed statistical error analysis of the parameter estimates. In particular, we studied the impact of different assumptions about the knowledge of the exact concentrations of the components, which in practice presents an experimental limitation for many systems. For example, the uncertainty in concentration may reflect imperfectly known extinction coefficients and stock concentrations or may account for different extents of partial inactivation when working with proteins at different pH values. Our results show that the global analysis can yield reliable estimates of the thermodynamic parameters for intrinsic binding and protonation, and that in the context of the global analysis the exact molecular component concentrations may not be required. Additionally, a comparison of data from different experimental strategies illustrates the benefit of conducting experiments at a range of temperatures.     

A structure-guided approach for protein pocket modeling and affinity prediction

Binding affinity prediction is frequently addressed using computational models constructed solely with molecular structure and activity data. We present a hybrid structure-guided strategy that combines molecular similarity, docking, and multiple-instance learning such that information from protein structures can be used to inform models of structure–activity relationships. The Surflex-QMOD approach has been shown to produce accurate predictions of binding affinity by constructing an interpretable physical model of a binding site with no experimental binding site structural information. We introduce a method to integrate protein structure information into the model induction process in order to construct more robust physical models. The structure-guided models accurately predict binding affinities over a broad range of compounds while producing more accurate representations of the protein pockets and ligand binding modes. Structure-guidance for the QMOD method yielded significant performance improvements, both for affinity and pose prediction, especially in cases where predictions were made on ligands very different from those used for model induction.     

Ensemble of linear models for predicting drug properties

We propose a new classification method for the prediction of drug properties, called random feature subset boosting for linear discriminant analysis (LDA). The main novelty of this method is the ability to overcome the problems with constructing ensembles of linear discriminant models based on generalized eigenvectors of covariance matrices. Such linear models are popular in building classification-based structure-activity relationships. The introduction of ensembles of LDA models allows for an analysis of more complex problems than by using single LDA, for example, those involving multiple mechanisms of action. Using four data sets, we show experimentally that the method is competitive with other recently studied chemoinformatic methods, including support vector machines and models based on decision trees. We present an easy scheme for interpreting the model despite its apparent sophistication. We also outline theoretical evidence as to why, contrary to the conventional AdaBoost ensemble algorithm, this method is able to increase the accuracy of LDA models.     

Performance of (consensus) kNN QSAR for predicting estrogenic activity in a large diverse set of organic compounds

A novel method (in the context of quantitative structure-activity relationship (QSAR)) based on the k nearest neighbour (kNN) principle, has recently been introduced for the derivation of predictive structure-activity relationships. Its performance has been tested for estimating the estrogen binding affinity of a diverse set of 142 organic molecules. Highly predictive models have been obtained. Moreover, it has been demonstrated that consensus-type kNN QSAR models, derived from the arithmetic mean of individual QSAR models were statistically robust and provided more accurate predictions than the great majority of the individual QSAR models. Finally, the consensus QSAR method was tested with 3D QSAR and log P data from a widely used steroid benchmark data set.     

Ligand and structure-based methodologies for the prediction of the activity of G protein-coupled receptor ligands

Accurate in silico models for the quantitative prediction of the activity of G protein-coupled receptor (GPCR) ligands would greatly facilitate the process of drug discovery and development. Several methodologies have been developed based on the properties of the ligands, the direct study of the receptor-ligand interactions, or a combination of both approaches. Ligand-based three-dimensional quantitative structure-activity relationships (3D-QSAR) techniques, not requiring knowledge of the receptor structure, have been historically the first to be applied to the prediction of the activity of GPCR ligands. They are generally endowed with robustness and good ranking ability; however they are highly dependent on training sets. Structure-based techniques generally do not provide the level of accuracy necessary to yield meaningful rankings when applied to GPCR homology models. However, they are essentially independent from training sets and have a sufficient level of accuracy to allow an effective discrimination between binders and nonbinders, thus qualifying as viable lead discovery tools. The combination of ligand and structure-based methodologies in the form of receptor-based 3D-QSAR and ligand and structure-based consensus models results in robust and accurate quantitative predictions. The contribution of the structure-based component to these combined approaches is expected to become more substantial and effective in the future, as more sophisticated scoring functions are developed and more detailed structural information on GPCRs is gathered.     

Virtual screening of integrase inhibitors by large scale binding free energy calculations: the SAMPL4 challenge

As part of the SAMPL4 blind challenge, filtered AutoDock Vina ligand docking predictions and large scale binding energy distribution analysis method binding free energy calculations have been applied to the virtual screening of a focused library of candidate binders to the LEDGF site of the HIV integrase protein. The computational protocol leveraged docking and high level atomistic models to improve enrichment. The enrichment factor of our blind predictions ranked best among all of the computational submissions, and second best overall. This work represents to our knowledge the first example of the application of an all-atom physics-based binding free energy model to large scale virtual screening. A total of 285 parallel Hamiltonian replica exchange molecular dynamics absolute protein-ligand binding free energy simulations were conducted starting from docked poses. The setup of the simulations was fully automated, calculations were distributed on multiple computing resources and were completed in a 6-weeks period. The accuracy of the docked poses and the inclusion of intramolecular strain and entropic losses in the binding free energy estimates were the major factors behind the success of the method. Lack of sufficient time and computing resources to investigate additional protonation states of the ligands was a major cause of mispredictions. The experiment demonstrated the applicability of binding free energy modeling to improve hit rates in challenging virtual screening of focused ligand libraries during lead optimization.     

Improved CoMFA modeling by optimization of settings

The possibility of improving the predictive ability of comparative molecular field analysis (CoMFA) by settings optimization has been evaluated to show that CoMFA predictive ability can be improved. Ten different CoMFA settings are evaluated, producing a total of 6120 models. This method has been applied to nine different data sets, including the widely used benchmark steroid data set, as well as eight other data sets proposed as QSAR benchmarking data sets by Sutherland et al. (J. Med. Chem. 2004, 47, 5541-5554). All data sets have been studied using training and test sets to allow for both internal (q(2)) and external (r(2)(pred)) predictive ability assessment. CoMFA settings optimization was successful in developing models with improved q(2) and r(2)(pred) as compared to default CoMFA modeling. Optimized CoMFA is compared with comparative molecular similarity indices analysis (CoMSIA) and holographic quantitative structure-activity relationship (HQSAR) models and found to consistently produce models with improved or equivalent q(2) and r(2)(pred). The ability of settings optimization to improve model predictive ability has been validated using both internal and external predictions, and the risk of chance correlation has been evaluated using response variable randomization tests.     

Multiple automatic base selection: Protein–ligand docking based on incremental construction without manual intervention

A possible way of tackling the molecular docking problem arising in computer- aided drug design is the use of the incremental construction method. This method consists of three steps: the selection of a part of a molecule, a so- called base fragment, the placement of the base fragment into the active site of a protein, and the subsequent reconstruction of the complete drug molecule. Assuming that a part of a drug molecule is known, which is specific enough to be a good base fragment, the method is proven to be successful for a large set of docking examples. In addition, it leads to the fastest algorithms for flexible docking published so far. In most real-world applications of docking, large sets of ligands have to be tested for affinity to a given protein. Thus, manual selection of a base fragment is not practical. On the other hand, the selection of a base fragment is critical in that only few selections lead to a low-energy structure. We overcome this limitation by selecting a representative set of base fragments instead of a single one. In this paper, we present a set of rules and algorithms to automate this selection. In addition, we extend the incremental construction method to deal with multiple fragmentations of the drug molecule. Our results show that with multiple automated base selection, the quality of the docking predictions is almost as good as with one manually preselected base fragment. In addition, the set of solutions is more diverse and alternative binding modes with low scores are found. Although the run time of the overall algorithm increases, the method remains fast enough to search through large ligand data sets.     

Structure--activity landscape index: identifying and quantifying activity cliffs

A new method for analyzing a structure-activity relationship is proposed. By use of a simple quantitative index, one can readily identify "structure-activity cliffs": pairs of molecules which are most similar but have the largest change in activity. We show how this provides a graphical representation of the entire SAR, in a way that allows the salient features of the SAR to be quickly grasped. In addition, the approach allows us view the SARs in a data set at different levels of detail. The method is tested on two data sets that highlight its ability to easily extract SAR information. Finally, we demonstrate that this method is robust using a variety of computational control experiments and discuss possible applications of this technique to QSAR model evaluation.     

Modeling data from titration,amide H/D exchange,and mass spectrometry to obtain protein-ligand binding constants

We recently reported a new method for quantification of protein-ligand interaction by mass spectrometry, titration and H/D exchange (PLIMSTEX) for determining the binding stoichiometry and affinity of a wide range of protein-ligand interactions. Here we describe the method for analyzing the PLIMSTEX titration curves and evaluate the effect of various models on the precision and accuracy for determining binding constants using H/D exchange and a titration. The titration data were fitted using a 1:n protein:ligand sequential binding model, where n is the number of binding sites for the same ligand. An ordinary differential equation was used for the first time in calculating the free ligand concentration from the total ligand concentration. A nonlinear least squares regression method was applied to minimize the error between the calculated and the experimentally measured deuterium shift by varying the unknown parameters. A resampling method and second-order statistics were used to evaluate the uncertainties of the fitting parameters. The interaction of intestinal fatty-acid-binding protein (IFABP) with a fatty-acid carboxylate and that of calmodulin with Ca(2+) are used as two tests. The modeling process described here not only is a new tool for analyzing H/D exchange data acquired by ESI-MS, but also possesses novel aspects in modeling experimental titration data to determine the affinity of ligand binding.     

Different binding modes of structurally diverse ligands for human D3DAR

Five different dopamine D3 receptors (D3DARs) models were created considering some suggested binding modes for D3DAR antagonists reported in earlier computational studies. Different hypotheses are justified because of the lack of experimental information about the putative site of interaction and are also due to the variability in scaffolds and size of D3DAR ligands. In this study 114 potent and selective D3DAR antagonists or partial agonists are used as key experimental information to discriminate the most reliable receptor model and to build a docking based 3D quantitative structure-activity relationship model able to indicate the ligand properties and the residues important for activity. The ability of this D3DAR model to discriminate the binding mode of different classes of ligands, showing a good quantitative correlation with their activity, encourages us to use it for screening novel lead compounds.