Quantitative Structure-Activity Relationship Modeling of Kinase Selectivity Profiles

Large-scale comparison of machine learning methods for profiling prediction of kinase inhibitors

Conventional machine learning (ML) and deep learning (DL) play a key role in the selectivity prediction of kinase inhibitors. A number of models based on available datasets can be used to predict the kinase profile of compounds, but there is still controversy about the advantages and disadvantages of ML and DL for such tasks. In this study, we constructed a comprehensive benchmark dataset of kinase inhibitors, involving in 141,086 unique compounds and 216,823 well-defined bioassay data points for 354 kinases. We then systematically compared the performance of 12 ML and DL methods on the kinase profiling prediction task. Extensive experimental results reveal that (1) Descriptor-based ML models generally slightly outperform fingerprint-based ML models in terms of predictive performance. RF as an ensemble learning approach displays the overall best predictive performance. (2) Single-task graph-based DL models are generally inferior to conventional descriptor- and fingerprint-based ML models, however, the corresponding multi-task models generally improves the average accuracy of kinase profile prediction. For example, the multi-task FP-GNN model outperforms the conventional descriptor- and fingerprint-based ML models with an average AUC of 0.807. (3) Fusion models based on voting and stacking methods can further improve the performance of the kinase profiling prediction task, specifically, RF::AtomPairs + FP2 + RDKitDes fusion model performs best with the highest average AUC value of 0.825 on the test sets. These findings provide useful information for guiding choices of the ML and DL methods for the kinase profiling prediction tasks. Finally, an online platform called KIPP ( https://kipp.idruglab.cn ) and python software are developed based on the best models to support the kinase profiling prediction, as well as various kinase inhibitor identification tasks including virtual screening, compound repositioning and target fishing.

JAK1 Pseudokinase V666G Mutant Dominantly Impairs JAK3 Phosphorylation and IL-2 Signaling

Overactive Janus kinases (JAKs) are known to drive leukemia, making them well-suited targets for treatment. We sought to identify new JAK-activating mutations and instead found a JAK1-inactivating pseudokinase mutation, V666G. In contrast to other pseudokinase mutations that canonically lead to an active kinase, the JAK1 V666G mutation led to under-activation seen by reduced phosphorylation. To understand the functional role of JAK1 V666G in modifying kinase activity we investigated its influence on other JAK kinases and within the Interleukin-2 pathway. JAK1 V666G not only inhibited its own activity, but its presence could inhibit other JAK kinases. These findings provide new insights into the potential of JAK1 pseudokinase to modulate its own activity, as well as of other JAK kinases. Thus, the features of the JAK1 V666 region in modifying JAK kinases can be exploited to allosterically inhibit overactive JAKs.

The Many Faces of JAKs and STATs Within the COVID-19 Storm

The positive-sense single stranded RNA virus, Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2), resulted in a global pandemic with horrendous health and economic consequences not seen in a century. At a finer scale, immunologically, many of these devastating effects by SARS-CoV-2 can be traced to a "cytokine storm" resulting in the simultaneous activation of Janus Kinases (JAKs) and Signal Transducers and Activators of Transcription (STAT) proteins downstream of the many cytokine receptor families triggered by elevated cytokines found in Coronavirus Disease 2019 (COVID-19). In this report, cytokines found in the storm are discussed in relation to the JAK-STAT pathway in response to SARS-CoV-2 and the lessons learned from RNA viruses and previous Coronaviruses (CoVs). Therapeutic strategies to counteract the SARS-CoV-2 mediated storm are discussed with an emphasis on cell signaling and JAK inhibition.

Computational Approaches for the Design of (Mutant-)Selective Tyrosine Kinase Inhibitors: State-of-the-Art and Future Prospects

Kinases remain one of the major attractive therapeutic targets for a large number of indications such as cancer, rheumatoid arthritis, cardiac failure and many others. Design and development of kinase inhibitors (ATP-competitive, allosteric or covalent) is a clinically validated and successful strategy in the pharmaceutical industry. The perks come with limitations, particularly the development of resistance to highly potent and selective inhibitors. When this happens, the cycle needs to be repeated, i.e., the design and development of kinase inhibitors active against the mutated forms. The complexity of tumor milieu makes it awfully difficult for these molecularly-targeted therapies to work. Every year newer and better versions of these agents are introduced in the clinic. Several computational approaches such as structure-, ligand-based or hybrid ones continue to live up to their potential in discovering novel kinase inhibitors. New schools of thought in this area continue to emerge, e.g., development of dual-target kinase inhibitors. But there are fundamental issues with this approach. It is indeed difficult to selectively optimize binding at two entirely different or related kinases. In addition to the conventional strategies, modern technologies (machine learning, deep learning, artificial intelligence, etc.) started yielding the results and building success stories. Computational tools invariably played a critical role in catalysing the phenomenal progress in kinase drug discovery field. The present review summarized the progress in utilizing computational methods and tools for discovering (mutant-)selective tyrosine kinase inhibitor drugs in the last three years (2017-2019). Representative investigations have been discussed, while others are merely listed. The author believes that the enthusiastic reader will be inspired to dig out the cited literature extensively to appreciate the progress made so far and the future prospects of the field.

Imputation of Assay Bioactivity Data Using Deep Learning

We describe a novel deep learning neural network method and its application to impute assay pIC₅₀ values. Unlike conventional machine learning approaches, this method is trained on sparse bioactivity data as input, typical of that found in public and commercial databases, enabling it to learn directly from correlations between activities measured in different assays. In two case studies on public domain data sets we show that the neural network method outperforms traditional quantitative structure-activity relationship (QSAR) models and other leading approaches. Furthermore, by focusing on only the most confident predictions the accuracy is increased to R² > 0.9 using our method, as compared to R² = 0.44 when reporting all predictions.

Application of Multivariate Adaptive Regression Splines (MARSplines) for Predicting Hansen Solubility Parameters Based on 1D and 2D Molecular Descriptors Computed from SMILES String

A new method of Hansen solubility parameters (HSPs) prediction was developed by combining the multivariate adaptive regression splines (MARSplines) methodology with a simple multivariable regression involving 1D and 2D PaDEL molecular descriptors. In order to adopt the MARSplines approach to QSPR/QSAR problems, several optimization procedures were proposed and tested. The effectiveness of the obtained models was checked via standard QSPR/QSAR internal validation procedures provided by the QSARINS software and by predicting the solubility classification of polymers and drug-like solid solutes in collections of solvents. By utilizing information derived only from SMILES strings, the obtained models allow for computing all of the three Hansen solubility parameters including dispersion, polarization, and hydrogen bonding. Although several descriptors are required for proper parameters estimation, the proposed procedure is simple and straightforward and does not require a molecular geometry optimization. The obtained HSP values are highly correlated with experimental data, and their application for solving solubility problems leads to essentially the same quality as for the original parameters. Based on provided models, it is possible to characterize any solvent and liquid solute for which HSP data are unavailable.