As part of the Community Structure-Activity Resource (CSAR) center, a set of 343 high-quality, protein–ligand crystal structures were assembled with experimentally determined Kd or Ki information from the literature. We encouraged the community to score the crystallographic poses of the complexes by any method of their choice. The goal of the exercise was to (1) evaluate the current ability of the field to predict activity from structure and (2) investigate the properties of the complexes and methods that appear to hinder scoring. A total of 19 different methods were submitted with numerous parameter variations for a total of 64 sets of scores from 16 participating groups. Linear regression and nonparametric tests were used to correlate scores to the experimental values. Correlation to experiment for the various methods ranged R2 = 0.58–0.12, Spearman ρ = 0.74–0.37, Kendall τ = 0.55–0.25, and median unsigned error = 1.00–1.68 pKd units. All types of scoring functions—force field based, knowledge based, and empirical—had examples with high and low correlation, showing no bias/advantage for any particular approach. The data across all the participants were combined to identify 63 complexes that were poorly scored across the majority of the scoring methods and 123 complexes that were scored well across the majority. The two sets were compared using a Wilcoxon rank-sum test to assess any significant difference in the distributions of >400 physicochemical properties of the ligands and the proteins. Poorly scored complexes were found to have ligands that were the same size as those in well-scored complexes, but hydrogen bonding and torsional strain were significantly different. These comparisons point to a need for CSAR to develop data sets of congeneric series with a range of hydrogen-bonding and hydrophobic characteristics and a range of rotatable bonds.

Recently some of us made intrinsic aqueous solubility measurements for 132 structurally diverse drugs and biologically significant molecules. We then issued, in conjunction with this Journal, an open prediction challenge in a paper where we reported the intrinsic solubilities for 100 of these compounds as a training set and listed the structures of the remaining 32 compounds which formed the prediction set. 1 This solubility challenge data were also made available online including a link on this Journal's Web site. The formal Solubility Challenge was held from approximately July 15, 2008 through September 15, 2008 although late and/or changed entries were accepted until the end of September. More than 100 entries to the Solubility Challenge have been received. In several cases multiple entries came from the same person or group. In addition, more than 5% of the prediction sheet entries were incomplete in that predictions were not reported for all 32 compounds of the prediction set. These incomplete entries were not included in the overall findings given here, but the submitted prediction sheets, like those of all other entries, were scored and returned to the contestants along with a copy of the findings of this challenge. Overall, 99 completed entries were scored and are reported here.Before presenting some of the findings from this challenge, it is important to reiterate what was stated at the outset of the challenge. The goal of this Solubility Challenge is not to identify a 'winner' but rather to adVance our general understanding of how to better perform aqueous solubility estimations. The findings of this challenge also proVide us a perspectiVe on the current state of predictiVe capabilities.The prediction set of compounds and their measured intrinsic solubilities are given in Table 1. Some qualifiers had to be applied to the prediction set to fairly and unambiguously score the contestants' data. Two of the prediction set compounds exhibit polymorph solubility behavior, and the solubility of each polymorph has been

A major goal in drug design is the improvement of computational methods for docking and scoring. The Community Structure Activity Resource (CSAR) has collected several data sets from industry and added in-house data sets that may be used for this purpose (). CSAR has currently obtained data from Abbott, GlaxoSmithKline, and Vertex and is working on obtaining data from several others. Combined with our in-house projects, we are providing a data set consisting of 6 protein targets, 647 compounds with biological affinities, and 82 crystal structures. Multiple congeneric series are available for several targets with a few representative crystal structures of each of the series. These series generally contain a few inactive compounds, usually not available in the literature, to provide an upper bound to the affinity range. The affinity ranges are typically 3–4 orders of magnitude per series. For our in-house projects, we have had compounds synthesized for biological testing. Affinities were measured by Thermofluor, Octet RED, and isothermal titration calorimetry for the most soluble. This allows the direct comparison of the biological affinities for those compounds, providing a measure of the variance in the experimental affinity. It appears that there can be considerable variance in the absolute value of the affinity, making the prediction of the absolute value ill-defined. However, the relative rankings within the methods are much better, and this fits with the observation that predicting relative ranking is a more tractable problem computationally. For those in-house compounds, we also have measured the following physical properties: logD, logP, thermodynamic solubility, and pKa. This data set also provides a substantial decoy set for each target consisting of diverse conformations covering the entire active site for all of the 58 CSAR-quality crystal structures. The CSAR data sets (CSAR-NRC HiQ and the 2012 release) provide substantial, publically available, curated data sets for use in parametrizing and validating docking and scoring methods.

There are several Quantitative Structure-Activity Relationship (QSAR) methods to assist in the design of compounds for medicinal use. Owing to the different QSAR methodologies, deciding which QSAR method to use depends on the composition of system of interest and the desired results. The relationship between a compound's binding affinity/activity to its structural properties was first noted in the 1930s by Hammett and later refined by Hansch and Fujita in the mid-1960s. In 1988 Cramer and coworkers created Comparative Molecular Field Analysis (CoMFA) incorporating the three-dimensional (3D) aspects of the compounds, specifically the electrostatic fields of the compound, into the QSAR model. Hopfinger and coworkers included an additional dimension to 3D-QSAR methodology in 1997 that eliminated the question of "Which conformation to use in a QSAR study?", creating 4D-QSAR. In 1999 Chemical Computing Group Inc. (CCG) developed the Binary-QSAR methodology and added novel 3D-QSAR descriptors to the traditional QSAR model allowing the 3D properties of compounds to be incorporated into the traditional QSAR model. Recently CCG released Probabilistic Receptor Potentials to calculate the substrate's atomic preferences in the active site. These potentials are constructed by fitting analytical functions to experimental properties of the substrates using knowledge-based methods. An overview of these and other QSAR methods will be discussed along with an in-depth examination of the methodologies used to construct QSAR models. Also, included in this chapter is a case study of molecules used to create QSAR models utilizing different methodologies and QSAR programs.

Blockage of the human ether-a-go-go related gene (hERG) potassium ion channel is a major factor related to cardiotoxicity. Hence, drugs binding to this channel have become an important biological end point in side effects screening. A set of 250 structurally diverse compounds screened for hERG activity from the literature was assembled using a set of reliability filters. This data set was used to construct a set of two-state hERG QSAR models. The descriptor pool used to construct the models consisted of 4D-fingerprints generated from the thermodynamic distribution of conformer states available to a molecule, 204 traditional 2D descriptors and 76 3D VolSurf-like descriptors computed using the Molecular Operating Environment (MOE) software. One model is a continuous partial least-squares (PLS) QSAR hERG binding model. Another related model is an optimized binary classification QSAR model that classifies compounds as active or inactive. This binary model achieves 91% accuracy over a large range of molecular diversity spanning the training set. Two external test sets were constructed. One test set is the condensed PubChem bioassay database containing 876 compounds, and the other test set consists of 106 additional compounds found in the literature. Both of the test sets were used to validate the binary QSAR model. The binary QSAR model permits a structural interpretation of possible sources for hERG activity. In particular, the presence of a polar negative group at a distance of 6-8 A from a hydrogen bond donor in a compound is predicted to be a quite structure-specific pharmacophore that increases hERG blockage. Since a data set of high chemical diversity was used to construct the binary model, it is applicable for performing general virtual hERG screening.

The traditional biological assay is very time-consuming, and thus the ability to quickly screen large numbers of compounds against a specific biological target is appealing. To speed up the biological evaluation of compounds, high-throughput screening is widely used in the fields of biomedical, biological information, and drug discovery. The research presented in this study focuses on the use of support vector machines, a machine learning method, various classes of molecular descriptors, and different sampling techniques to overcome overfitting to classify compounds for cytotoxicity with respect to the Jurkat cell line. The cell cytotoxicity data set is imbalanced (a few active compounds and very many inactive compounds), and the ability of the predictive modeling methods is adversely affected in these situations. Commonly imbalanced data sets are overfit with respect to the dominant classified end point; in this study the models routinely overfit toward inactive (noncytotoxic) compounds when the imbalance was substantial. Support vector machine (SVM) models were used to probe the proficiency of different classes of molecular descriptors and oversampling ratios. The SVM models were constructed from 4D-FPs, MOE (1D, 2D, and 21/2D), noNP+MOE, and CATS2D trial descriptors pools and compared to the predictive abilities of CATS2D-based random forest models. Compared to previous results in the literature, the SVM models built from oversampled data sets exhibited better predictive abilities for the training and external test sets.

Little attention has been given to the selection of trial descriptor sets when designing a QSAR analysis even though a great number of descriptor classes, and often a greater number of descriptors within a given class, are now available. This paper reports an effort to explore interrelationships between QSAR models and descriptor sets. Zhou and co-workers (Zhou et al., Nano Lett. 2008, 8 (3), 859-865) designed, synthesized, and tested a combinatorial library of 80 surface modified, that is decorated, multi-walled carbon nanotubes for their composite nanotoxicity using six endpoints all based on a common 0 to 100 activity scale. Each of the six endpoints for the 29 most nanotoxic decorated nanotubes were incorporated as the training set for this study. The study reported here includes trial descriptor sets for all possible combinations of MOE, VolSurf, and 4D-fingerprints (FP) descriptor classes, as well as including and excluding explicit spatial contributions from the nanotube. Optimized QSAR models were constructed from these multiple trial descriptor sets. It was found that (a) both the form and quality of the best QSAR models for each of the endpoints are distinct and (b) some endpoints are quite dependent upon 4D-FP descriptors of the entire nanotube-decorator complex. However, other endpoints yielded equally good models only using decorator descriptors with and without the decorator-only 4D-FP descriptors. Lastly, and most importantly, the quality, significance, and interpretation of a QSAR model were found to be critically dependent on the trial descriptor sets used within a given QSAR endpoint study.

The human ether-a-go-go related gene (hERG) potassium ion channel plays a key role in cardiotoxicity and is therefore a key target as part of preclinical drug discovery toxicity screening. The PubChem hERG Bioassay data set, composed of 1668 compounds, was used to construct an in silico screening model. The corresponding trial models were constructed from a descriptor pool composed of 4D fingerprints (4D-FP) and traditional 2D and 3D VolSurf-like molecular descriptors. A final binary classification model was constructed via a support vector machine (SVM). The resultant model was then validated using the PubChem hERG Bioassay data set (AID 376) and an external hERG data set by evaluating the model's ability to determine hERG blockers from nonblockers. The external data set (the test set) consisted of 356 compounds collected from available literature data and consisting of 287 actives and 69 inactives. Four different sampling protocols and a 10-fold cross-correlation analysis--used in the validation process to evaluate classification models--explored the impact of the active--inactive data imbalance distribution of the PubChem high-throughput data set. Four different data sets were explored, and the one employing Lipinski's rule-of-five coupled with measures of relative molecular lipophilicity performed the best in the 10-fold cross-correlation validation of the training data set as well as overall prediction accuracy of the external test sets. The linear SVM binary classification model building strategy was applied to different combinations of MOE (traditional 2D, "21/2D", and 3D VolSurf-like) and 4D-FP molecular descriptors to further explore and refine previously proposed key descriptors, identify new significant features that contribute to the prediction of hERG toxicity, and construct the optimal SVM binary classification model from a shrunken descriptor pool. The accuracy, sensitivity, and specificity of the best model determined from 10-fold cross-validation are 95, 90, and 96%, respectively; the overall accuracy is near 87% for the external set. The models constructed in this study demonstrate the following: (i) robustness based upon performance in accuracy across the structural diversity of the training set, (ii) ability to predict a compound's "predisposition" to block hERG ion channels, and (iii) define and illustrate structural features that can be overlaid onto the chemical structures to aid in the 3D structure-activity interpretation of the hERG blocking effect.