Deep generative design with 3D pharmacophoric constraints

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 10× 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.

A novel deep generative model combines convolution and graph neural networks to allow 3D-aware molecular design. We show how 3D pharmacophoric information can be incorporated into generative models and apply our model to both linker and R-group design.     

Beyond generative models: superfast traversal,optimization, novelty,exploration and discovery (STONED) algorithm for molecules using SELFIES

Inverse design allows the generation of molecules with desirable physical quantities using property optimization. Deep generative models have recently been applied to tackle inverse design, as they possess the ability to optimize molecular properties directly through structure modification using gradients. While the ability to carry out direct property optimizations is promising, the use of generative deep learning models to solve practical problems requires large amounts of data and is very time-consuming. In this work, we propose STONED – a simple and efficient algorithm to perform interpolation and exploration in the chemical space, comparable to deep generative models. STONED bypasses the need for large amounts of data and training times by using string modifications in the SELFIES molecular representation. First, we achieve non-trivial performance on typical benchmarks for generative models without any training. Additionally, we demonstrate applications in high-throughput virtual screening for the design of drugs, photovoltaics, and the construction of chemical paths, allowing for both property and structure-based interpolation in the chemical space. Overall, we anticipate our results to be a stepping stone for developing more sophisticated inverse design models and benchmarking tools, ultimately helping generative models achieve wider adoption.

Interpolation and exploration within the chemical space for inverse design.     

Structure-based de novo drug design using 3D deep generative models

Deep generative models are attracting much attention in the field of de novo molecule design. Compared to traditional methods, deep generative models can be trained in a fully data-driven way with little requirement for expert knowledge. Although many models have been developed to generate 1D and 2D molecular structures, 3D molecule generation is less explored, and the direct design of drug-like molecules inside target binding sites remains challenging. In this work, we introduce DeepLigBuilder, a novel deep learning-based method for de novo drug design that generates 3D molecular structures in the binding sites of target proteins. We first developed Ligand Neural Network (L-Net), a novel graph generative model for the end-to-end design of chemically and conformationally valid 3D molecules with high drug-likeness. Then, we combined L-Net with Monte Carlo tree search to perform structure-based de novo drug design tasks. In the case study of inhibitor design for the main protease of SARS-CoV-2, DeepLigBuilder suggested a list of drug-like compounds with novel chemical structures, high predicted affinity, and similar binding features to those of known inhibitors. The current version of L-Net was trained on drug-like compounds from ChEMBL, which could be easily extended to other molecular datasets with desired properties based on users'' demands and applied in functional molecule generation. Merging deep generative models with atomic-level interaction evaluation, DeepLigBuilder provides a state-of-the-art model for structure-based de novo drug design and lead optimization.

DeepLigBuilder, a novel deep generative model for structure-based de novo drug design, directly generates 3D structures of drug-like compounds in the target binding site.     

A molecular approach to rationally constructing specific fluorogenic substrates for the detection of acetylcholinesterase activity in live cells,mice brains and tissues

Acetylcholinesterase (AChE) is an extremely critical hydrolase tightly associated with neurological diseases. Currently, developing specific substrates for imaging AChE activity still remains a great challenge due to the interference from butyrylcholinesterase (BChE) and carboxylesterase (CE). Herein, we propose an approach to designing specific substrates for AChE detection by combining dimethylcarbamate choline with a self-immolative scaffold. The representative P10 can effectively eliminate the interference from CE and BChE. The high specificity of P10 has been proved via imaging AChE activity in cells. Moreover, P10 can also be used to successfully map AChE activity in different regions of a normal mouse brain, which may provide important data for AChE evaluation in clinical studies. Such a rational and effective approach can also provide a solid basis for designing probes with different properties to study AChE in biosystems and another way to design specific substrates for other enzymes.

In this work, a new approach was developed for designing the representative P10 with high selectivity and sensitivity for imaging AChE activity in the cells and normal mouse brain.     

Attention-based generative models for de novo molecular design

Attention mechanisms have led to many breakthroughs in sequential data modeling but have yet to be incorporated into any generative algorithms for molecular design. Here we explore the impact of adding self-attention layers to generative β-VAE models and show that those with attention are able to learn a complex “molecular grammar” while improving performance on downstream tasks such as accurately sampling from the latent space (“model memory”) or exploring novel chemistries not present in the training data. There is a notable relationship between a model''s architecture, the structure of its latent memory and its performance during inference. We demonstrate that there is an unavoidable tradeoff between model exploration and validity that is a function of the complexity of the latent memory. However, novel sampling schemes may be used that optimize this tradeoff. We anticipate that attention will play an important role in future molecular design algorithms that can make efficient use of the detailed molecular substructures learned by the transformer.

An implementation of attention within the variational autoencoder framework for continuous representation of molecules. The addition of attention significantly increases model performance for complex tasks such as exploration of novel chemistries.     

Graph neural network based coarse-grained mapping prediction

The selection of coarse-grained (CG) mapping operators is a critical step for CG molecular dynamics (MD) simulation. It is still an open question about what is optimal for this choice and there is a need for theory. The current state-of-the art method is mapping operators manually selected by experts. In this work, we demonstrate an automated approach by viewing this problem as supervised learning where we seek to reproduce the mapping operators produced by experts. We present a graph neural network based CG mapping predictor called Deep Supervised Graph Partitioning Model (DSGPM) that treats mapping operators as a graph segmentation problem. DSGPM is trained on a novel dataset, Human-annotated Mappings (HAM), consisting of 1180 molecules with expert annotated mapping operators. HAM can be used to facilitate further research in this area. Our model uses a novel metric learning objective to produce high-quality atomic features that are used in spectral clustering. The results show that the DSGPM outperforms state-of-the-art methods in the field of graph segmentation. Finally, we find that predicted CG mapping operators indeed result in good CG MD models when used in simulation.

We propose a scalable graph neural network-based method for automating coarse-grained mapping prediction for molecules.     

Small molecule-mediated co-assembly of amyloid-β oligomers reduces neurotoxicity through promoting non-fibrillar aggregation

Amyloid-β (Aβ) oligomers, particularly low molecular weight (LMW) oligomers, rather than fibrils, contribute very significantly to the onset and progression of Alzheimer''s Disease (AD). However, due to the inherent heterogeneity and metastability of oligomers, most of the conventional anti-oligomer therapies have indirectly modulated oligomers'' toxicity through manipulating Aβ self-assembly to reduce oligomer levels, which are prone to suffering from the risk of regenerating toxic oligomers from the products of modulation. To circumvent this disadvantage, we demonstrate, for the first time, rational design of rigid pincer-like scaffold-based small molecules with blood–brain barrier permeability that specifically co-assemble with LMW Aβ oligomers through directly binding to the exposed hydrophobic regions of oligomers to form non-fibrillar, degradable, non-toxic co-aggregates. As a proof of concept, treatment with a europium complex (EC) in such a structural mode can rescue Aβ-mediated dysfunction in C. elegans models of AD in vivo. This small molecule-mediated oligomer co-assembly strategy offers an efficient approach for AD treatment.

A rational design of pincer-like scaffold-based small molecule with blood-brain barrier permeability that can specifically co-assemble with low molecular weight Aβ oligomers to form non-fibrillar, degradable, non-toxic co-aggregates.     

A chemically consistent graph architecture for massive reaction networks applied to solid-electrolyte interphase formation

Modeling reactivity with chemical reaction networks could yield fundamental mechanistic understanding that would expedite the development of processes and technologies for energy storage, medicine, catalysis, and more. Thus far, reaction networks have been limited in size by chemically inconsistent graph representations of multi-reactant reactions (e.g. A + B → C) that cannot enforce stoichiometric constraints, precluding the use of optimized shortest-path algorithms. Here, we report a chemically consistent graph architecture that overcomes these limitations using a novel multi-reactant representation and iterative cost-solving procedure. Our approach enables the identification of all low-cost pathways to desired products in massive reaction networks containing reactions of any stoichiometry, allowing for the investigation of vastly more complex systems than previously possible. Leveraging our architecture, we construct the first ever electrochemical reaction network from first-principles thermodynamic calculations to describe the formation of the Li-ion solid electrolyte interphase (SEI), which is critical for passivation of the negative electrode. Using this network comprised of nearly 6000 species and 4.5 million reactions, we interrogate the formation of a key SEI component, lithium ethylene dicarbonate. We automatically identify previously proposed mechanisms as well as multiple novel pathways containing counter-intuitive reactions that have not, to our knowledge, been reported in the literature. We envision that our framework and data-driven methodology will facilitate efforts to engineer the composition-related properties of the SEI – or of any complex chemical process – through selective control of reactivity.

A chemically consistent graph architecture enables autonomous identification of novel solid-electrolyte interphase formation pathways from a massive reaction network.     

A pyrone remodeling strategy to access diverse heterocycles: application to the synthesis of fascaplysin natural products

The synthesis of diverse N-fused heterocycles, including the pyrido[1,2-a]indole scaffold, using an efficient pyrone remodeling strategy is described. The pyrido[1,2-a]indole core was demonstrated to be a versatile scaffold that can be site-selectively functionalized. The utility of this novel annulation strategy was showcased in a concise formal synthesis of three fascaplysin congeners.

The synthesis of diverse N-fused heterocycles, including the pyrido[1,2-a]indole scaffold, using an efficient pyrone remodeling strategy is described.     

Retrosynthetic accessibility score (RAscore) – rapid machine learned synthesizability classification from AI driven retrosynthetic planning

Computer aided synthesis planning (CASP) is part of a suite of artificial intelligence (AI) based tools that are able to propose synthesis routes to a wide range of compounds. However, at present they are too slow to be used to screen the synthetic feasibility of millions of generated or enumerated compounds before identification of potential bioactivity by virtual screening (VS) workflows. Herein we report a machine learning (ML) based method capable of classifying whether a synthetic route can be identified for a particular compound or not by the CASP tool AiZynthFinder. The resulting ML models return a retrosynthetic accessibility score (RAscore) of any molecule of interest, and computes at least 4500 times faster than retrosynthetic analysis performed by the underlying CASP tool. The RAscore should be useful for pre-screening millions of virtual molecules from enumerated databases or generative models for synthetic accessibility and produce higher quality databases for virtual screening of biological activity.

The retrosynthetic accessibility score (RAscore) is based on AI driven retrosynthetic planning, and is useful for rapid scoring of synthetic feasability and pre-screening of large datasets of virtual/generated molecules.     

Machine-enabled inverse design of inorganic solid materials: promises and challenges

Developing high-performance advanced materials requires a deeper insight and search into the chemical space. Until recently, exploration of materials space using chemical intuitions built upon existing materials has been the general strategy, but this direct design approach is often time and resource consuming and poses a significant bottleneck to solve the materials challenges of future sustainability in a timely manner. To accelerate this conventional design process, inverse design, which outputs materials with pre-defined target properties, has emerged as a significant materials informatics platform in recent years by leveraging hidden knowledge obtained from materials data. Here, we summarize the latest progress in machine-enabled inverse materials design categorized into three strategies: high-throughput virtual screening, global optimization, and generative models. We analyze challenges for each approach and discuss gaps to be bridged for further accelerated and rational data-driven materials design.

The grand challenge of materials science, discovery of novel materials with target properties, can be greatly accelerated by machine-learned inverse design strategies.     

BonDNet: a graph neural network for the prediction of bond dissociation energies for charged molecules

A broad collection of technologies, including e.g. drug metabolism, biofuel combustion, photochemical decontamination of water, and interfacial passivation in energy production/storage systems rely on chemical processes that involve bond-breaking molecular reactions. In this context, a fundamental thermodynamic property of interest is the bond dissociation energy (BDE) which measures the strength of a chemical bond. Fast and accurate prediction of BDEs for arbitrary molecules would lay the groundwork for data-driven projections of complex reaction cascades and hence a deeper understanding of these critical chemical processes and, ultimately, how to reverse design them. In this paper, we propose a chemically inspired graph neural network machine learning model, BonDNet, for the rapid and accurate prediction of BDEs. BonDNet maps the difference between the molecular representations of the reactants and products to the reaction BDE. Because of the use of this difference representation and the introduction of global features, including molecular charge, it is the first machine learning model capable of predicting both homolytic and heterolytic BDEs for molecules of any charge. To test the model, we have constructed a dataset of both homolytic and heterolytic BDEs for neutral and charged (−1 and +1) molecules. BonDNet achieves a mean absolute error (MAE) of 0.022 eV for unseen test data, significantly below chemical accuracy (0.043 eV). Besides the ability to handle complex bond dissociation reactions that no previous model could consider, BonDNet distinguishes itself even in only predicting homolytic BDEs for neutral molecules; it achieves an MAE of 0.020 eV on the PubChem BDE dataset, a 20% improvement over the previous best performing model. We gain additional insight into the model''s predictions by analyzing the patterns in the features representing the molecules and the bond dissociation reactions, which are qualitatively consistent with chemical rules and intuition. BonDNet is just one application of our general approach to representing and learning chemical reactivity, and it could be easily extended to the prediction of other reaction properties in the future.

Prediction of bond dissociation energies for charged molecules with a graph neural network enabled by global molecular features and reaction difference features between products and reactants.     

Strain visualization for strained macrocycles

Strain has a unique and sometimes unpredictable impact on the properties and reactivity of molecules. To thoroughly describe strain in molecules, a computational tool that relates strain energy to reactivity by localizing and quantifying strain was developed. Strain energy is calculated local to every coordinate in the molecule and areas of higher strain are shown experimentally to be more reactive. Not only does this tool directly compare strain energy in parts of the same molecule, but it also computes total strain to give a full picture of molecular strain energy. It is freely available to the public on GitHub under the name StrainViz and much of the workflow is automated to simplify use for non-experts. Unique insight into the reactivity of curved aromatic molecules and strained alkyne bioorthogonal reagents is described within.

Strain has a unique and sometimes unpredictable impact on the properties and reactivity of molecules.     

Carbene-catalyzed enantioselective annulation of dinucleophilic hydrazones and bromoenals for access to aryl-dihydropyridazinones and related drugs

4,5-Dihydropyridazinones bearing an aryl substituent at the C6-position are important motifs in drug molecules. Herein, we developed an efficient protocol to access aryl-dihydropyridazinone molecules via carbene-catalyzed asymmetric annulation between dinucleophilic arylidene hydrazones and bromoenals. Key steps in this reaction include polarity-inversion of aryl aldehyde-derived hydrazones followed by chemo-selective reaction with enal-derived α,β-unsaturated acyl azolium intermediates. The aryl-dihydropyridazinone products accessed by our protocol can be readily transformed into drugs and bioactive molecules.

Polarity inversion of arylidene hydrazones to react with bromoenals via carbene organic catalysis is disclosed. The reaction enantioselectively affords 6-aryl-4,5-dihydropyridazinones and related drugs with proven commercial applications.     

Phosphoryl- and phosphonium-bridged viologens as stable two- and three-electron acceptors for organic electrodes

Low molecular weight organic molecules that can accept multiple electrons at high reduction potentials are sought after as electrode materials for high-energy sustainable batteries. To date their synthesis has been difficult, and organic scaffolds for electron donors significantly outnumber electron acceptors. Herein, we report the synthesis and electronic properties of two highly electron-deficient phosphaviologen derivatives from a phosphorus-bridged 4,4''-bipyridine and characterize their electrochemical properties. Phosphaviologen sulfide (PVS) and P-methyl phosphaviologen (PVM) accept two and three electrons at high reduction potentials, respectively. PVM can reversibly accept three electrons between 3–3.6 V vs. Li/Li⁺ with an equivalent molecular weight of 102 g (mol⁻¹ e⁻) (262 mA h g⁻¹), making it a promising scaffold for sustainable organic electrode materials having high specific energy densities.

Two strongly electron-accepting viologens, including an intriguing tricationic species, are reported. The utility of the tricationic viologen for energy storage has been showcased via use as electrode in a proof-of-concept battery.     

Is it time for biocatalysis in fragment-based drug discovery?

The use of biocatalysts for fragment-based drug discovery has yet to be fully investigated, despite the promise enzymes hold for the synthesis of poly-functional, non-protected small molecules. Here we analyze products of the biocatalysis literature to demonstrate the potential for not only fragment generation, but also the enzyme-mediated elaboration of these fragments. Our analysis demonstrates that biocatalytic products can readily populate 3D chemical space, offering diverse catalytic approaches to help generate new, bioactive molecules.

This perspective discusses how biocatalysis could play an important role in the future fragment-based drug discovery.     

Target-templated de novo design of macrocyclic d-/l-peptides: discovery of drug-like inhibitors of PD-1

Peptides are a rapidly growing class of therapeutics with various advantages over traditional small molecules, especially for targeting difficult protein–protein interactions. However, current structure-based methods are largely limited to natural peptides and are not suitable for designing bioactive cyclic topologies that go beyond natural l-amino acids. Here, we report a generalizable framework that exploits the computational power of Rosetta, in terms of large-scale backbone sampling, side-chain composition and energy scoring, to design heterochiral cyclic peptides that bind to a protein surface of interest. To showcase the applicability of our approach, we developed two new inhibitors (PD-i3 and PD-i6) of programmed cell death 1 (PD-1), a key immune checkpoint in oncology. A comprehensive biophysical evaluation was performed to assess their binding to PD-1 as well as their blocking effect on the endogenous PD-1/PD-L1 interaction. Finally, NMR elucidation of their in-solution structures confirmed our de novo design approach.

In silico design of heterochiral cyclic peptides that bind to a specific surface patch on the target protein (PD-1, in this case) and disrupt protein–protein interactions.     

Access to P-chiral sec- and tert-phosphine oxides enabled by Le-Phos-catalyzed asymmetric kinetic resolution

The synthesis of P-stereogenic building blocks is extremely difficult. Herein we report an efficient kinetic resolution of secondary phosphine oxides via a Le-Phos-catalyzed asymmetric allylation reaction with Morita–Baylis–Hillman carbonates. This method provides facile access to enantioenriched secondary and tertiary P-chiral phosphine oxides with broad substrate scope, both of which could serve as P-stereogenic synthons, and can be rapidly incorporated into a given scaffold bearing a P-stereocenter. The highly desirable late stage modifications demonstrate the practicability of our method and can be a critical contribution to obtaining optimal P-chiral catalysts and ligands.

Herein we report an efficient kinetic resolution of secondary phosphine oxides via a Le-Phos-catalyzed asymmetric allylation reaction with Morita–Baylis–Hillman carbonates.     

Machine learning dielectric screening for the simulation of excited state properties of molecules and materials

Accurate and efficient calculations of absorption spectra of molecules and materials are essential for the understanding and rational design of broad classes of systems. Solving the Bethe–Salpeter equation (BSE) for electron–hole pairs usually yields accurate predictions of absorption spectra, but it is computationally expensive, especially if thermal averages of spectra computed for multiple configurations are required. We present a method based on machine learning to evaluate a key quantity entering the definition of absorption spectra: the dielectric screening. We show that our approach yields a model for the screening that is transferable between multiple configurations sampled during first principles molecular dynamics simulations; hence it leads to a substantial improvement in the efficiency of calculations of finite temperature spectra. We obtained computational gains of one to two orders of magnitude for systems with 50 to 500 atoms, including liquids, solids, nanostructures, and solid/liquid interfaces. Importantly, the models of dielectric screening derived here may be used not only in the solution of the BSE but also in developing functionals for time-dependent density functional theory (TDDFT) calculations of homogeneous and heterogeneous systems. Overall, our work provides a strategy to combine machine learning with electronic structure calculations to accelerate first principles simulations of excited-state properties.

Machine learning can circumvent explicit calculation of dielectric response in first principles methods and accelerate simulations of optical properties of complex materials at finite temperature.