The number of protein kinase inhibitors (PKIs) approved worldwide continues to grow steadily, with 39 drugs approved in the period between 2001 and January 2018. PKIs on the market have been the subject of many reviews, and structure-property relationships specific to this class of drugs have been inferred. However, the large number of PKIs under development is often overlooked. In this paper, we present PKIDB (Protein Kinase Inhibitor Database), a monthly-updated database gathering approved PKIs as well as PKIs currently in clinical trials. The database compiles currently 180 inhibitors ranging from phase 0 to 4 clinical trials along with annotations extracted from seven public resources. The distribution and property ranges of standard physicochemical properties are presented. They can be used as filters to better prioritize compound selection for future screening campaigns. Interestingly, more than one-third of the kinase inhibitors violate at least one Lipinski’s rule. A Principal Component Analysis (PCA) reveals that Type-II inhibitors are mapped to a distinct chemical space as compared to orally administrated drugs as well as to other types of kinase inhibitors. Using a Principal Moment of Inertia (PMI) analysis, we show that PKIs under development tend to explore new shape territories as compared to approved PKIs. In order to facilitate the analysis of the protein space, the kinome tree has been annotated with all protein kinases being targeted by PKIs. Finally, we analyzed the pipeline of the pharmaceutical companies having PKIs on the market or still under development. We hope that this work will assist researchers in the kinase field in identifying and designing the next generation of kinase inhibitors for still untargeted kinases. The PKIDB database is freely accessible from a website at http://www.icoa.fr/pkidb and can be easily browsed through a user-friendly spreadsheet-like interface.

Since the first approval of a protein kinase inhibitor (PKI) by the Food and Drug Administration (FDA) in 2001, 55 new PKIs have reached the market, and many inhibitors are currently being evaluated in clinical trials. This is a clear indication that protein kinases still represent major drug targets for the pharmaceutical industry. In a previous work, we have introduced PKIDB, a publicly available database, gathering PKIs that have already been approved (Phase 4), as well as those currently in clinical trials (Phases 0 to 3). This database is updated frequently, and an analysis of the new data is presented here. In addition, we compared the set of PKIs present in PKIDB with the PKIs in early preclinical studies found in ChEMBL, the largest publicly available chemical database. For each dataset, the distribution of physicochemical descriptors related to drug-likeness is presented. From these results, updated guidelines to prioritize compounds for targeting protein kinases are proposed. The results of a principal component analysis (PCA) show that the PKIDB dataset is fully encompassed within all PKIs found in the public database. This observation is reinforced by a principal moments of inertia (PMI) analysis of all molecules. Interestingly, we notice that PKIs in clinical trials tend to explore new 3D chemical space. While a great majority of PKIs is located on the area of “flatland”, we find few compounds exploring the 3D structural space. Finally, a scaffold diversity analysis of the two datasets, based on frequency counts was performed. The results give insight into the chemical space of PKIs, and can guide researchers to reach out new unexplored areas. PKIDB is freely accessible from the following website: http://www.icoa.fr/pkidb.

The accurate prediction of molecular properties, such as lipophilicity and aqueous solubility, are of great importance and pose challenges in several stages of the drug discovery pipeline. Machine learning methods, such as graph-based neural networks (GNNs), have shown exceptionally good performance in predicting these properties. In this work, we introduce a novel GNN architecture, called directed edge graph isomorphism network (D-GIN). It is composed of two distinct sub-architectures (D-MPNN, GIN) and achieves an improvement in accuracy over its sub-architectures employing various learning, and featurization strategies. We argue that combining models with different key aspects help make graph neural networks deeper and simultaneously increase their predictive power. Furthermore, we address current limitations in assessment of deep-learning models, namely, comparison of single training run performance metrics, and offer a more robust solution.

As the application of computational methods in drug discovery pipelines becomes more widespread it is increasingly important to understand how reproducible their results are and how sensitive they are to choices made in simulation setup and analysis. Here we use ensemble simulation protocols, termed ESMACS (enhanced sampling of molecular dynamics with approximation of continuum solvent), to investigate the sensitivity of the popular molecular mechanics Poisson-Boltzmann surface area (MMPBSA) methodology. Using the bromodomain-containing protein 4 (BRD4) system bound to a diverse set of ligands as our target, we show that robust rankings can be produced only through combining ensemble sampling with multiple trajectories and enhanced solvation via an explicit ligand hydration shell.

Over the past two decades, the use of fragment‐based lead generation has become a common, mature approach to identify tractable starting points in chemical space for the drug discovery process. This approach naturally involves the study of the binding properties of highly heterogeneous ligands. Such datasets challenge computational techniques to provide comparable binding free energy estimates from different binding modes. The performance of a range of statistically robust ensemble‐based binding free energy calculation protocols, called ESMACS (enhanced sampling of molecular dynamics with approximation of continuum solvent), is evaluated. Ligands designed to target two binding pockets in the lactate dehydogenase, a target protein, which vary in size, charge, and binding mode, are studied. When compared to experimental results, excellent statistical rankings are obtained across this highly diverse set of ligands. In addition, three approaches to account for entropic contributions are investigated: 1) normal mode analysis, 2) weighted solvent accessible surface area (WSAS), and 3) variational entropy. Normal mode analysis and WSAS correlate strongly with each other—although the latter is computationally far cheaper—but do not improve rankings. Variational entropy corrects exaggerated discrimination of ligands bound in different pockets but creates three outliers which reduce the quality of the overall ranking.

Fragment-based drug design is an established routine approach in both experimental and computational spheres. Growing fragment hits into viable ligands has increasingly shifted into the spotlight. FastGrow is an application based on a shape search algorithm that addresses this challenge at high speeds of a few milliseconds per fragment. It further features a pharmacophoric interaction description, ensemble flexibility, as well as geometry optimization to become a fully fledged structure-based modeling tool. All features were evaluated in detail on a previously reported collection of fragment growing scenarios extracted from crystallographic data. FastGrow was also shown to perform competitively versus established docking software. A case study on the DYRK1A kinase, using recently reported new chemotypes, illustrates FastGrow’s features in practice and its ability to identify active fragments. FastGrow is freely available to the public as a web server at https://fastgrow.plus/ and is part of the SeeSAR 3D software package.