Confident application of a global human liver microsomal activity QSAR.

Metabolic stability is an important property of drug candidates and pharmaceutical companies often have human liver microsomal (HLM) data for a large number of molecules, enabling development of global quantitative structure-activity relationship models. RESULTS: This study describes a strategy for building a global HLM quantitative structure-activity relationship model, applicable also to datasets of limited size. By using external congeneric test sets, a realistic description of the performance in the future applied setting and a reliable prediction confidence method is obtained. CONCLUSION: The limited ability of the HLM model to generalize in chemical space to show the importance of internal model development and continuous updating of global HLM models, as well as the importance of a validated prediction confidence method.

A hierarchical screening approach to enantiomeric separation.

The screening of a number of chiral stationary phases (CSPs) with different modifiers in supercritical fluid chromatography to find a chromatographic method for separation of enantiomers can be time-consuming. Computational methods for data analysis were utilized to establish a hierarchical screening strategy, using a dataset of 110 drug-like chiral compounds with diverse structures tested on 15 CSPs with two different modifiers. This dataset was analyzed using a combinatorial algorithm, principal component analysis (PCA), and a correlation matrix. The primary goal was to find a set of eight columns resolving a large number of compounds, but also having complementary enantioselective properties. In addition to the hereby defined hierarchical experimental strategy, quantitative structure enantioselective models (QSERs) were evaluated. The diverse chemical space and relatively limited size of the training set reduced the accuracy of the QSERs. However, including separation factors from other CSPs increased the accuracies of the QSERs substantially. Hence, such combined models can support the experimental strategy in prioritizing the CSPs of the second screening phase, when a compound is not separated by the primary set of columns.

Deconvoluting Kinase Inhibitor Induced Cardiotoxicity.

Many drugs designed to inhibit kinases have their clinical utility limited by cardiotoxicity-related label warnings or prescribing restrictions. While this liability is widely recognized, designing safer kinase inhibitors (KI) requires knowledge of the causative kinase(s). Efforts to unravel the kinases have encountered pharmacology with nearly prohibitive complexity. At therapeutically relevant concentrations, KIs show promiscuity distributed across the kinome. Here, to overcome this complexity, 65 KIs with known kinome-scale polypharmacology profiles were assessed for effects on cardiomyocyte (CM) beating. Changes in human iPSC-CM beat rate and amplitude were measured using label-free cellular impedance. Correlations between beat effects and kinase inhibition profiles were mined by computation analysis (Matthews Correlation Coefficient) to identify associated kinases. Thirty kinases met criteria of having (1) pharmacological inhibition correlated with CM beat changes, (2) expression in both human-induced pluripotent stem cell-derived cardiomyocytes and adult heart tissue, and (3) effects on CM beating following single gene knockdown. A subset of these 30 kinases were selected for mechanistic follow up. Examples of kinases regulating processes spanning the excitation-contraction cascade were identified, including calcium flux (RPS6KA3, IKBKE) and action potential duration (MAP4K2). Finally, a simple model was created to predict functional cardiotoxicity whereby inactivity at three sentinel kinases (RPS6KB1, FAK, STK35) showed exceptional accuracy in vitro and translated to clinical KI safety data. For drug discovery, identifying causative kinases and introducing a predictive model should transform the ability to design safer KI medicines. For cardiovascular biology, discovering kinases previously unrecognized as influencing cardiovascular biology should stimulate investigation of underappreciated signaling pathways.

Exploring in silico prediction of the unbound brain-to-plasma drug concentration ratio: model validation, renewal, and interpretation.

Recently, we built an in silico model to predict the unbound brain-to-plasma concentration ratio (Kp,uu,brain), a measure of the distribution of a compound between the blood plasma and the brain. Here, we validate the previous model with new additional data points expanding the chemical space and use that data also to renew the model. The model building process was similar to our previous approach; however, a new set of descriptors, molecular signatures, was included to facilitate the model interpretation from a structure perspective. The best consensus model shows better predictive power than the previous model (R(2) = 0.6 vs. R(2) = 0.53, when the same 99 compounds were used as test set). The two-class classification accuracy increased from 76% using the previous model to 81%. Furthermore, the atom-summarized gradient based on molecular signature descriptors was proposed as an interesting new approach to interpret the Kp,uu,brain machine learning model and scrutinize structure Kp,uu,brain relationships for investigated compounds.

Assessment of machine learning reliability methods for quantifying the applicability domain of QSAR regression models.

The vastness of chemical space and the relatively small coverage by experimental data recording molecular properties require us to identify subspaces, or domains, for which we can confidently apply QSAR models. The prediction of QSAR models in these domains is reliable, and potential subsequent investigations of such compounds would find that the predictions closely match the experimental values. Standard approaches in QSAR assume that predictions are more reliable for compounds that are "similar" to those in subspaces with denser experimental data. Here, we report on a study of an alternative set of techniques recently proposed in the machine learning community. These methods quantify prediction confidence through estimation of the prediction error at the point of interest. Our study includes 20 public QSAR data sets with continuous response and assesses the quality of 10 reliability scoring methods by observing their correlation with prediction error. We show that these new alternative approaches can outperform standard reliability scores that rely only on similarity to compounds in the training set. The results also indicate that the quality of reliability scoring methods is sensitive to data set characteristics and to the regression method used in QSAR. We demonstrate that at the cost of increased computational complexity these dependencies can be leveraged by integration of scores from various reliability estimation approaches. The reliability estimation techniques described in this paper have been implemented in an open source add-on package ( https://bitbucket.org/biolab/orange-reliability ) to the Orange data mining suite.

Localized heuristic inverse quantitative structure activity relationship with bulk descriptors using numerical gradients.

State-of-the-art quantitative structure-activity relationship (QSAR) models are often based on nonlinear machine learning algorithms, which are difficult to interpret. From a pharmaceutical perspective, QSARs are used to enhance the chemical design process. Ultimately, they should not only provide a prediction but also contribute to a mechanistic understanding and guide modifications to the chemical structure, promoting compounds with desirable biological activity profiles. Global ranking of descriptor importance and inverse QSAR have been used for these purposes. This paper introduces localized heuristic inverse QSAR, which provides an assessment of the relative ability of the descriptors to influence the biological response in an area localized around the predicted compound. The method is based on numerical gradients with parameters optimized using data sets sampled from analytical functions. The heuristic character of the method reduces the computational requirements and makes it applicable not only to fragment based methods but also to QSARs based on bulk descriptors. The application of the method is illustrated on congeneric QSAR data sets, and it is shown that the predicted influential descriptors can be used to guide structural modifications that affect the biological response in the desired direction. The method is implemented into the AZOrange Open Source QSAR package. The current implementation of localized heuristic inverse QSAR is a step toward a generally applicable method for elucidating the structure activity relationship specifically for a congeneric region of chemical space when using QSARs based on bulk properties. Consequently, this method could contribute to accelerating the chemical design process in pharmaceutical projects, as well as provide information that could enhance the mechanistic understanding for individual scaffolds.

QSAR with experimental and predictive distributions: an information theoretic approach for assessing model quality.

We propose that quantitative structure-activity relationship (QSAR) predictions should be explicitly represented as predictive (probability) distributions. If both predictions and experimental measurements are treated as probability distributions, the quality of a set of predictive distributions output by a model can be assessed with Kullback-Leibler (KL) divergence: a widely used information theoretic measure of the distance between two probability distributions. We have assessed a range of different machine learning algorithms and error estimation methods for producing predictive distributions with an analysis against three of AstraZeneca's global DMPK datasets. Using the KL-divergence framework, we have identified a few combinations of algorithms that produce accurate and valid compound-specific predictive distributions. These methods use reliability indices to assign predictive distributions to the predictions output by QSAR models so that reliable predictions have tight distributions and vice versa. Finally we show how valid predictive distributions can be used to estimate the probability that a test compound has properties that hit single- or multi- objective target profiles.

Similarity boosted quantitative structure-activity relationship--a systematic study of enhancing structural descriptors by molecular similarity.

The concept of molecular similarity is one of the most central in the fields of predictive toxicology and quantitative structure-activity relationship (QSAR) research. Many toxicological responses result from a multimechanistic process and, consequently, structural diversity among the active compounds is likely. Combining this knowledge, we introduce similarity boosted QSAR modeling, where we calculate molecular descriptors using similarities with respect to representative reference compounds to aid a statistical learning algorithm in distinguishing between different structural classes. We present three approaches for the selection of reference compounds, one by literature search and two by clustering. Our experimental evaluation on seven publicly available data sets shows that the similarity descriptors used on their own perform quite well compared to structural descriptors. We show that the combination of similarity and structural descriptors enhances the performance and that a simple stacking approach is able to use the complementary information encoded by the different descriptor sets to further improve predictive results. All software necessary for our experiments is available within the cheminformatics software framework AZOrange.

AZOrange - High performance open source machine learning for QSAR modeling in a graphical programming environment.

BACKGROUND: Machine learning has a vast range of applications. In particular, advanced machine learning methods are routinely and increasingly used in quantitative structure activity relationship (QSAR) modeling. QSAR data sets often encompass tens of thousands of compounds and the size of proprietary, as well as public data sets, is rapidly growing. Hence, there is a demand for computationally efficient machine learning algorithms, easily available to researchers without extensive machine learning knowledge. In granting the scientific principles of transparency and reproducibility, Open Source solutions are increasingly acknowledged by regulatory authorities. Thus, an Open Source state-of-the-art high performance machine learning platform, interfacing multiple, customized machine learning algorithms for both graphical programming and scripting, to be used for large scale development of QSAR models of regulatory quality, is of great value to the QSAR community. RESULTS: This paper describes the implementation of the Open Source machine learning package AZOrange. AZOrange is specially developed to support batch generation of QSAR models in providing the full work flow of QSAR modeling, from descriptor calculation to automated model building, validation and selection. The automated work flow relies upon the customization of the machine learning algorithms and a generalized, automated model hyper-parameter selection process. Several high performance machine learning algorithms are interfaced for efficient data set specific selection of the statistical method, promoting model accuracy. Using the high performance machine learning algorithms of AZOrange does not require programming knowledge as flexible applications can be created, not only at a scripting level, but also in a graphical programming environment. CONCLUSIONS: AZOrange is a step towards meeting the needs for an Open Source high performance machine learning platform, supporting the efficient development of highly accurate QSAR models fulfilling regulatory requirements.