Computational methods for analysis and inference of kinase/inhibitor relationships

Poor Generalization by Current Deep Learning Models for Predicting Binding Affinities of Kinase Inhibitors

The extreme surge of interest over the past decade surrounding the use of neural networks has inspired many groups to deploy them for predicting binding affinities of drug-like molecules to their receptors. A model that can accurately make such predictions has the potential to screen large chemical libraries and help streamline the drug discovery process. However, despite reports of models that accurately predict quantitative inhibition using protein kinase sequences and inhibitors' SMILES strings, it is still unclear whether these models can generalize to previously unseen data. Here, we build a Convolutional Neural Network (CNN) analogous to those previously reported and evaluate the model over four datasets commonly used for inhibitor/kinase predictions. We find that the model performs comparably to those previously reported, provided that the individual data points are randomly split between the training set and the test set. However, model performance is dramatically deteriorated when all data for a given inhibitor is placed together in the same training/testing fold, implying that information leakage underlies the models' performance. Through comparison to simple models in which the SMILES strings are tokenized, or in which test set predictions are simply copied from the closest training set data points, we demonstrate that there is essentially no generalization whatsoever in this model. In other words, the model has not learned anything about molecular interactions, and does not provide any benefit over much simpler and more transparent models. These observations strongly point to the need for richer structure-based encodings, to obtain useful prospective predictions of not-yet-synthesized candidate inhibitors.

Kinome-wide identification of phosphorylation networks in eukaryotic proteomes

Motivation

Signaling and metabolic pathways are finely regulated by a network of protein phosphorylation events. Unraveling the nature of this intricate network, composed of kinases, target proteins and their interactions, is therefore of crucial importance. Although thousands of kinase-specific phosphorylations (KsP) have been annotated in model organisms their kinase-target network is far from being complete, with less studied organisms lagging behind.

Results

In this work, we achieved an automated and accurate identification of kinase domains, inferring the residues that most likely contribute to peptide specificity. We integrated this information with the target peptides of known human KsP to predict kinase-specific interactions in other eukaryotes through a deep neural network, outperforming similar methods. We analyzed the differential conservation of kinase specificity among eukaryotes revealing the high conservation of the specificity of tyrosine kinases. With this approach we discovered 1590 novel KsP of potential clinical relevance in the human proteome.

Availability and implementation

http://akid.bio.uniroma2.it.

Supplementary information

Supplementary data are available at Bioinformatics online.

Development of Certain Protein Kinase Inhibitors with the Components from Traditional Chinese Medicine

Traditional Chinese medicines (TCMs) have been used in China for more than two thousand years, and some of them have been confirmed to be effective in cancer treatment. Protein kinases play critical roles in control of cell growth, proliferation, migration, survival, and angiogenesis and mediate their biological effects through their catalytic activity. In recent years, numerous protein kinase inhibitors have been developed and are being used clinically. Anticancer TCMs represent a large class of bioactive substances, and some of them display anticancer activity via inhibiting protein kinases to affect the phosphoinositide 3-kinase, serine/threonine-specific protein kinases, pechanistic target of rapamycin (PI3K/AKT/mTOR), P38, mitogen-activated protein kinase (MAPK) and extracellular signal-regulated kinases (ERK) pathways. In the present article, we comprehensively reviewed several components isolated from anticancer TCMs that exhibited significantly inhibitory activity toward a range of protein kinases. These components, which belong to diverse structural classes, are reviewed herein, based upon the kinases that they inhibit. The prospects and problems in development of the anticancer TCMs are also discussed.

Profiling Prediction of Kinase Inhibitors: Toward the Virtual Assay

Kinome-wide screening would have the advantage of providing structure-activity relationships against hundreds of targets simultaneously. Here, we report the generation of ligand-based activity prediction models for over 280 kinases by employing Machine Learning methods on an extensive data set of proprietary bioactivity data combined with open data. High quality (AUC > 0.7) was achieved for ∼200 kinases by (1) combining open with proprietary data, (2) choosing Random Forest over alternative tested Machine Learning methods, and (3) balancing the training data sets. Tests on left-out and external data indicate a high value for virtual screening projects. Importantly, the derived models are evenly distributed across the kinome tree, allowing reliable profiling prediction for all kinase branches. The prediction quality was further improved by employing experimental bioactivity fingerprints of a small kinase subset. Overall, the generated models can support various hit identification tasks, including virtual screening, compound repurposing, and the detection of potential off-targets.

Toxicology Strategies for Drug Discovery: Present and Future

Attrition due to nonclinical safety represents a major issue for the productivity of pharmaceutical research and development (R&D) organizations, especially during the compound optimization stages of drug discovery and the early stages of clinical development. Focusing on decreasing nonclinical safety-related attrition is not a new concept, and various approaches have been experimented with over the last two decades. Front-loading testing funnels in Discovery with in vitro toxicity assays designed to rapidly identify unfavorable molecules was the approach adopted by most pharmaceutical R&D organizations a few years ago. However, this approach has also a non-negligible opportunity cost. Hence, significant refinements to the "fail early, fail often" paradigm have been proposed recently to reflect the complexity of accurately categorizing compounds with early data points without taking into account other important contextual aspects, in particular efficacious systemic and tissue exposures. This review provides an overview of toxicology approaches and models that can be used in pharmaceutical Discovery at the series/lead identification and lead optimization stages to guide and inform chemistry efforts, as well as a personal view on how to best use them to meet nonclinical safety-related attrition objectives consistent with a sustainable pharmaceutical R&D model. The scope of this review is limited to small molecules, as large molecules are associated with challenges that are quite different. Finally, a perspective on how several emerging technologies may impact toxicity evaluation is also provided.