Deep Generative Design with 3D Pharmacophoric Constraints

Abstract

Generative models have increasingly been proposed as a solution to the molecular design problem. However, it has proved challenging to control the design process or incorporate prior knowledge, limiting their practical use in drug discovery. In particular, generative methods have made limited use of three-dimensional (3D) structural information even though this is critical to binding. This work describes a method to incorporate such information and demonstrates the benefit of doing so. We combine an existing graph-based deep generative model, DeLinker, with a convolutional neural network to utilise physically-meaningful 3D representations of molecules and target pharmacophores. We apply our model, DEVELOP, to both linker and R-group design, demonstrating its suitability for both hit-to-lead and lead optimisation. The 3D pharmacophoric information results in improved generation and allows greater control of the design process. In multiple large-scale evaluations, we show that including 3D pharmacophoric constraints results in substantial improvements in the quality of generated molecules. On a challenging test set derived from PDBbind, our model improves the proportion of generated molecules with high 3D similarity to the original molecule by over 300%. In addition, DEVELOP recovers 10x more of the original molecules compared to the baseline DeLinker method. Our approach is general-purpose, readily modifiable to alternate 3D representations, and can be incorporated into other generative frameworks. Code is available at https://github.com/oxpig/DEVELOP.

An open-source molecular builder and free energy preparation workflow

“…Inspired by the BOMB approach to molecular design 13 , we grow from a fixed ligand core in order to maximise the use of binding mode information from structural biology sources, and rely on the user's medicinal chemistry expertise to suggest functional groups that improve binding affinity whilst remaining synthetically tractable. Alternative, generative methods for fragment growth 11,12 could be incorporated in future, but testing of expert medicinal chemist designs still remains popular today and FEgrow aims to automate this process.…”

“…The physics-based molecular mechanics-generalised Born with surface area (MM-GBSA) was then used to provide a more accurate score. Further, more recent, examples include FragExplorer 9 , which aims to grow or replace fragments to optimise molecular interaction fields generated by the GRID software 10 , DeepFrag 11 , which predicts appropriate fragment additions using a deep convolutional neural network trained on thousands of known protein-ligand complexes, and DEVELOP 12 , which uses deep generative models to output 3D molecules conditional on provided phamacophoric features of the binding site. However, the employed approximate physics-or knowledge-based approaches to scoring the designs will limit to some extent their ability to predict and optimise binding affinity.…”

PGMG: A Pharmacophore-Guided Deep Learning Approach for Bioactive Molecular Generation

“…However, the structure-activity relationship between the biological activity and the molecules generated by such methods is ambiguous. DeLinker 19 and SyntaLinker 20 adopt fragment-based drug design and retain active fragments while updating linkers to generate active molecules, and DEVELOP 17 combines DeLinker with chemical features as constraints to improve the quality of the generated molecules. The fragment-based approaches require explicit knowledge of the active fragments, which lead to a limited chemical space for the model.…”

“…For example, the choice of ligand-based drug design or structure-based drug design depends on what information can be used, which narrows down the application scenarios of many methods. It is clear that incorporating expert knowledge in the generation process is beneficial to the full utilization of bioactivity data information 17 . Therefore, combining deep generative models with knowledge in biochemistry to efficiently use the scarce data to design biologically active molecules is a crucial project.…”

The use of a graph database is a complementary approach to a classical similarity search for identifying commercially available fragment merges

“…De novo design using generative models has become increasingly common in the eld of molecular optimization, particularly for fragment growing [30][31][32][33][34][35] and linking [36][37][38][39]. However, there are no such existing models for performing fragment merging, partly owing to the lack of data and difculty in generating synthetic data for training.…”