Search

Kinase inhibitors are an important class of anticancer drugs, with 80 inhibitors clinically approved and >100 in active clinical testing. Most bind competitively in the ATP-binding site, leading to challenges with selectivity for a specific kinase, resulting in risks for toxicity and general off-target effects. Assessing the binding of an inhibitor for the entire kinome is experimentally possible but expensive. A reliable and interpretable computational prediction of kinase selectivity would greatly benefit the inhibitor discovery and optimization process. Here, we use machine learning on docked poses to address this need. To this end, we aggregated all known inhibitor-kinase affinities and generated the complete accompanying 3D interactome by docking all inhibitors to the respective high-quality X-ray structures. We then used this resource to train a neural network as a kinase-specific scoring function, which achieved an overall performance ( R 2 ) of 0.63-0.74 on unseen inhibitors across the kinome. The entire pipeline from molecule to 3D-based affinity prediction has been fully automated and wrapped in a freely available package. This has a graphical user interface that is tightly integrated with PyMOL to allow immediate adoption in the medicinal chemistry practice.

Recent advancements in deep learning and generative models have significantly expanded the applications of virtual screening for drug-like compounds. Here, we introduce a multitarget transformer model, PCMol , that leverages the latent protein embeddings derived from AlphaFold2 as a means of conditioning a de novo generative model on different targets. Incorporating rich protein representations allows the model to capture their structural relationships, enabling the chemical space interpolation of active compounds and target-side generalization to new proteins based on embedding similarities. In this work, we benchmark against other existing target-conditioned transformer models to illustrate the validity of using AlphaFold protein representations over raw amino acid sequences. We show that low-dimensional projections of these protein embeddings cluster appropriately based on target families and that model performance declines when these representations are intentionally corrupted. We also show that the PCMol model generates diverse, potentially active molecules for a wide array of proteins, including those with sparse ligand bioactivity data. The generated compounds display higher similarity known active ligands of held-out targets and have comparable molecular docking scores while maintaining novelty. Additionally, we demonstrate the important role of data augmentation in bolstering the performance of generative models in low-data regimes. Software package and AlphaFold protein embeddings are freely available at https://github.com/CDDLeiden/PCMol.

The interpretation of high-dimensional structure-activity data sets in drug discovery to predict ligand-protein interaction landscapes is a challenging task. Here we present Drug Discovery Maps (DDM), a machine learning model that maps the activity profile of compounds across an entire protein family, as illustrated here for the kinase family. DDM is based on the t-distributed stochastic neighbor embedding (t-SNE) algorithm to generate a visualization of molecular and biological similarity. DDM maps chemical and target space and predicts the activities of novel kinase inhibitors across the kinome. The model was validated using independent data sets and in a prospective experimental setting, where DDM predicted new inhibitors for FMS-like tyrosine kinase 3 (FLT3), a therapeutic target for the treatment of acute myeloid leukemia. Compounds were resynthesized, yielding highly potent, cellularly active FLT3 inhibitors. Biochemical assays confirmed most of the predicted off-targets. DDM is further unique in that it is completely open-source and available as a ready-to-use executable to facilitate broad and easy adoption.