Large Chemical Space Research Articles

AABBA Graph Kernel: Atom-Atom, Bond-Bond, and Bond-Atom Autocorrelations for Machine Learning.

Graphs are one of the most natural and powerful representations available for molecules; natural because they have an intuitive correspondence to skeletal formulas, the language used by chemists worldwide, and powerful, because they are highly expressive both globally (molecular topology) and locally (atom and bond properties). Graph kernels are used to transform molecular graphs into fixed-length vectors, which, based on their capacity of measuring similarity, can be used as fingerprints for machine learning (ML). To date, graph kernels have mostly focused on the atomic nodes of the graph. In this work, we developed a graph kernel based on atom-atom, bond-bond, and bond-atom (AABBA) autocorrelations. The resulting vector representations were tested on regression ML tasks on a data set of transition metal complexes; a benchmark motivated by the higher complexity of these compounds relative to organic molecules. In particular, we tested different flavors of the AABBA kernel in the prediction of the energy barriers and bond distances of the Vaska's complex data set (Friederich et al., Chem. Sci., 2020, 11, 4584). For a variety of ML models, including neural networks, gradient boosting machines, and Gaussian processes, we showed that AABBA outperforms the baseline including only atom-atom autocorrelations. Dimensionality reduction studies also showed that the bond-bond and bond-atom autocorrelations yield many of the most relevant features. We believe that the AABBA graph kernel can accelerate the exploration of large chemical spaces and inspire novel molecular representations in which both atomic and bond properties play an important role.

Development of Calcium-Dependent Phospholipase A2 Inhibitors to Target Cellular Senescence and Oxidative Stress in Neurodegenerative Diseases.

Significance: Cellular senescence is a critical process underlying aging and is associated with age-related diseases such as Alzheimer's disease. Lipids are implicated in cellular senescence. Fatty acids, particularly eicosanoids, have been associated with various forms of senescence and inflammation, and the associated reactive oxygen species production has been proposed as a therapeutic target for mitigating senescence. When overactivated, calcium-dependent phospholipase A2 (cPLA2) catalyzes the conversion of arachidonic acid into eicosanoids such as leukotrienes and prostaglandins. Recent Advances: With a growing understanding of the importance of lipids as mediators and modulators of senescence, cPLA2 has emerged as a compelling drug target. cPLA2 overactivation plays a significant role in several pathways associated with senescence, including neuroinflammation and oxidative stress. Critical Issues: Previous cPLA2 inhibitors have shown potential in ameliorating inflammation and oxidative stress, but the dominant hurdles in the central nervous system-targeting drug discovery are specificity and blood-brain barrier penetrance. Future Directions: With the need for more effective drugs against neurological diseases, we emphasize the significance of discovering new brain-penetrant, potent, and specific cPLA2 inhibitors. We discuss how the recently developed Virtual Synthon Hierarchical Enumeration Screening, an iterative synthon-based approach for fast structure-based virtual screening of billions of compounds, provides an efficient exploration of large chemical spaces for the discovery of brain-penetrant cPLA2 small-molecule inhibitors. Antioxid. Redox Signal. 00, 000-000.

Leveraging infrared spectroscopy for automated structure elucidation

The application of machine learning models in chemistry has made remarkable strides in recent years. While analytical chemistry has received considerable interest from machine learning practitioners, its adoption into everyday use remains limited. Among the available analytical methods, Infrared (IR) spectroscopy stands out in terms of affordability, simplicity, and accessibility. However, its use has been limited to the identification of a selected few functional groups, as most peaks lie beyond human interpretation. We present a transformer model that enables chemists to leverage the complete information contained within an IR spectrum to directly predict the molecular structure. To cover a large chemical space, we pretrain the model using 634,585 simulated IR spectra and fine-tune it on 3,453 experimental spectra. Our approach achieves a top–1 accuracy of 44.4% and top–10 accuracy of 69.8% on compounds containing 6 to 13 heavy atoms. When solely predicting scaffolds, the model accurately predicts the top–1 scaffold in 84.5% and among the top–10 in 93.0% of cases.

Active Learning Framework for Expediting the Search of Thermodynamically Stable MXenes in the Extensive Chemical Space.

MXenes possess a wide range of materials properties owing to their compositional and stoichiometric diversities, facilitating their utilization in various technological applications such as electrodes, catalysts, and supercapacitors. To explore their applicability, identification of thermodynamically stable and synthesizable MXenes should precede. The energy above the convex hull (Ehull) calculated using the density functional theory (DFT) is a powerful scale to probe the thermodynamic stability. However, the high calculation cost of DFT limits the search space of unknown chemistry. To address this challenge, this study proposes an active learning (AL) framework consisting of a surrogate model and utility function for expeditious identification of thermodynamically stable MXenes in the extensive chemical space of 23,857 MXenes with compositional and stoichiometric diversity. Exploiting the fast inference speed and the capability of the AL framework to accurately identify stable MXenes, only 480 DFT calculations were required to identify 126 thermodynamically stable MXenes; among these, the stabilities of 89 MXenes have not been previously reported. In contrast, only two stable MXenes were identified among randomly selected 1693 MXenes, demonstrating the inefficiency of using only DFT calculations in exploring a large chemical space. The AL framework successfully minimized the number of DFT calculations while maximizing that of thermodynamically stable MXenes identified and can contribute to future studies in finding stable MXenes expeditiously.

Identification of novel and potent inhibitors of SARS-CoV-2 main protease from DNA-encoded chemical libraries.

In vitro screening of large compound libraries with automated high-throughput screening is expensive and time-consuming and requires dedicated infrastructures. Conversely, the selection of DNA-encoded chemical libraries (DECLs) can be rapidly performed with routine equipment available in most laboratories. In this study, we identified novel inhibitors of SARS-CoV-2 main protease (Mpro) through the affinity-based selection of the DELopen library (open access for academics), containing 4.2 billion compounds. The identified inhibitors were peptide-like compounds containing an N-terminal electrophilic group able to form a covalent bond with the nucleophilic Cys145 of Mpro, as confirmed by x-ray crystallography. This DECL selection campaign enabled the discovery of the unoptimized compound SLL11 (IC50 = 30 nM), proving that the rapid exploration of large chemical spaces enabled by DECL technology allows for the direct identification of potent inhibitors avoiding several rounds of iterative medicinal chemistry. As demonstrated further by x-ray crystallography, SLL11 was found to adopt a highly unique U-shaped binding conformation, which allows the N-terminal electrophilic group to loop back to the S1' subsite while the C-terminal amino acid sits in the S1 subsite. MP1, a close analog of SLL11, showed antiviral activity against SARS-CoV-2 in the low micromolar range when tested in Caco-2 and Calu-3 (EC50 = 2.3 µM) cell lines. As peptide-like compounds can suffer from low cell permeability and metabolic stability, the cyclization of the compounds will be explored in the future to improve their antiviral activity.

Differentiable modeling and optimization of non-aqueous Li-based battery electrolyte solutions using geometric deep learning

Electrolytes play a critical role in designing next-generation battery systems, by allowing efficient ion transfer, preventing charge transfer, and stabilizing electrode-electrolyte interfaces. In this work, we develop a differentiable geometric deep learning (GDL) model for chemical mixtures, DiffMix, which is applied in guiding robotic experimentation and optimization towards fast-charging battery electrolytes. In particular, we extend mixture thermodynamic and transport laws by creating GDL-learnable physical coefficients. We evaluate our model with mixture thermodynamics and ion transport properties, where we show improved prediction accuracy and model robustness of DiffMix than its purely data-driven variants. Furthermore, with a robotic experimentation setup, Clio, we improve ionic conductivity of electrolytes by over 18.8% within 10 experimental steps, via differentiable optimization built on DiffMix gradients. By combining GDL, mixture physics laws, and robotic experimentation, DiffMix expands the predictive modeling methods for chemical mixtures and enables efficient optimization in large chemical spaces.

Identification of Novel Tyrosinase Inhibitors with Nanomolar Potency Using Virtual Screening Approaches.

Hyperpigmentation disorders are caused by excess production of the pigment melanin, catalyzed by the enzyme tyrosinase. Novel tyrosinase inhibitors are needed as therapeutic agents to treat these conditions. To discover new inhibitors, we performed a virtual screening of the ZINC20 library containing 1.4 billion compounds. An initial filter for drug-likeness, ADMET properties, and synthetic accessibility reduced the library to 10,217 hits. Quantitative structure-activity relationship (QSAR) modeling of this subset predicted nanomolar inhibitory potency for several chemical scaffolds. Comparative molecular docking studies and rigorous binding energy calculations further prioritized four cysteine-containing dipeptide compounds based on predicted strong binding affinity and mode to tyrosinase. Microsecond-long molecular dynamics simulations provided additional atomistic insights into the stability of inhibitor-enzyme binding interactions. This integrated computational workflow effectively sampled an extremely large chemical space to discover four novel tyrosinase inhibitors with half-maximal inhibitory concentration values below 10 nM. Overall, this demonstrates the power of virtual screening and multi-faceted computational techniques to accelerate the discovery of potent bioactive ligands from massive compound libraries by efficiently sampling chemical space.

Oxadiazolines as Photoreleasable Labels for Drug Target Identification.

Photoaffinity labeling is a widely used technique for studying ligand-protein and protein-protein interactions. Traditional photoaffinity labels utilize nonspecific C-H bond insertion reactions mediated by a highly reactive intermediate. Despite being the most widely used photoaffinity labels, diazirines exhibit limited compatibility with downstream organic reactions and suffer from storage stability concerns. This study introduces oxadiazolines as innovative and complementary photoactivatable labels for addition to the toolbox and demonstrates their application in vitro and through in cellulo labeling experiments. Oxadiazolines can be easily synthesized from ketone moieties and cleaved with 302-330 nm light to cleanly liberate a diazo reactive moiety that can covalently modify nucleophilic amino acid residues. Notably, oxadiazolines are compatible with various organic reaction conditions and functional groups, allowing for the exploration of a large chemical space. Several known inhibitors featuring the oxadiazoline functionality were prepared without affecting their binding affinity. Furthermore, we confirmed the ability of oxadiazolines to form covalent bonds with proteins upon UV-irradiation, both in vitro and in cellulo, yielding comparable results to those of the matched diazirine compounds.

Challenges Reconciling Theory and Experiments in the Prediction of Lattice Thermal Conductivity: The Case of Cu-Based Sulvanites.

The exploration of large chemical spaces in search of new thermoelectric materials requires the integration of experiments, theory, simulations, and data science. The development of high-throughput strategies that combine DFT calculations with machine learning has emerged as a powerful approach to discovering new materials. However, experimental validation is crucial to confirm the accuracy of these workflows. This validation becomes especially important in understanding the transport properties that govern the thermoelectric performance of materials since they are highly influenced by synthetic, processing, and operating conditions. In this work, we explore the thermal conductivity of Cu-based sulvanites by using a combination of theoretical and experimental methods. Previous discrepancies and significant variations in reported data for Cu3VS4 and Cu3VSe4 are explained using the Boltzmann Transport Equation for phonons and by synthesizing well-characterized defect-free samples. The use of machine learning approaches for extracting high-order force constants opens doors to charting the lattice thermal conductivity across the entire Cu-based sulvanite family-finding not only materials with κ l values below 2 W m-1 K-1 at moderate temperatures but also rationalizing their thermal transport properties based on chemical composition.

Enabling target-aware molecule generation to follow multi objectives with Pareto MCTS

Target-aware drug discovery has greatly accelerated the drug discovery process to design small-molecule ligands with high binding affinity to disease-related protein targets. Conditioned on targeted proteins, previous works utilize various kinds of deep generative models and have shown great potential in generating molecules with strong protein-ligand binding interactions. However, beyond binding affinity, effective drug molecules must manifest other essential properties such as high drug-likeness, which are not explicitly addressed by current target-aware generative methods. In this article, aiming to bridge the gap of multi-objective target-aware molecule generation in the field of deep learning-based drug discovery, we propose ParetoDrug, a Pareto Monte Carlo Tree Search (MCTS) generation algorithm. ParetoDrug searches molecules on the Pareto Front in chemical space using MCTS to enable synchronous optimization of multiple properties. Specifically, ParetoDrug utilizes pretrained atom-by-atom autoregressive generative models for the exploration guidance to desired molecules during MCTS searching. Besides, when selecting the next atom symbol, a scheme named ParetoPUCT is proposed to balance exploration and exploitation. Benchmark experiments and case studies demonstrate that ParetoDrug is highly effective in traversing the large and complex chemical space to discover novel compounds with satisfactory binding affinities and drug-like properties for various multi-objective target-aware drug discovery tasks.

Massively parallel analysis of single-molecule dynamics on next-generation sequencing chips.

Single-molecule techniques are ideally poised to characterize complex dynamics but are typically limited to investigating a small number of different samples. However, a large sequence or chemical space often needs to be explored to derive a comprehensive understanding of complex biological processes. Here we describe multiplexed single-molecule characterization at the library scale (MUSCLE), a method that combines single-molecule fluorescence microscopy with next-generation sequencing to enable highly multiplexed observations of complex dynamics. We comprehensively profiled the sequence dependence of DNA hairpin properties and Cas9-induced target DNA unwinding-rewinding dynamics. The ability to explore a large sequence space for Cas9 allowed us to identify a number of target sequences with unexpected behaviors. We envision that MUSCLE will enable the mechanistic exploration of many fundamental biological processes.

Identifying Synergistic Components of Botanical Fungicide Formulations Using Interpretable Graph Neural Networks.

Botanical formulations are promising candidates for developing new biopesticides that can protect crops from pests and diseases while reducing harm to the environment. These biopesticides can be combined with permeation enhancer compounds to boost their efficacy against pests and fungal diseases. However, finding synergistic combinations of these compounds is challenging due to the large and complex chemical space. In this paper, we propose a novel deep learning method that can predict the synergy of botanical products and permeation enhancers based on in vitro assay data. Our method uses a weighted combination of component feature vectors to represent the input mixtures, which enables the model to handle a variable number of components and to interpret the contribution of each component to the synergy. We also employ an ensemble of interpretation methods to provide insights into the underlying mechanisms of synergy. We validate our method by testing the predicted synergistic combinations in wet-lab experiments and show that our method can discover novel and effective biopesticides that would otherwise be difficult to find. Our method is generalizable and applicable to other domains, where predicting mixtures of chemical compounds is important.

Accelerated Discovery of Halide Perovskite Materials via Computational Methods: A Review.

Halide perovskites have gained considerable attention in materials science due to their exceptional optoelectronic properties, including high absorption coefficients, excellent charge-carrier mobilities, and tunable band gaps, which make them highly promising for applications in photovoltaics, light-emitting diodes, synapses, and other optoelectronic devices. However, challenges such as long-term stability and lead toxicity hinder large-scale commercialization. Computational methods have become essential in this field, providing insights into material properties, enabling the efficient screening of large chemical spaces, and accelerating discovery processes through high-throughput screening and machine learning techniques. This review further discusses the role of computational tools in the accelerated discovery of high-performance halide perovskite materials, like the double perovskites A2BX6 and A2BB'X6, zero-dimensional perovskite A3B2X9, and novel halide perovskite ABX6. This review provides significant insights into how computational methods have accelerated the discovery of high-performance halide perovskite. Challenges and future perspectives are also presented to stimulate further research progress.

ADMET-AI: a machine learning ADMET platform for evaluation of large-scale chemical libraries.

The emergence of large chemical repositories and combinatorial chemical spaces, coupled with high-throughput docking and generative AI, have greatly expanded the chemical diversity of small molecules for drug discovery. Selecting compounds for experimental validation requires filtering these molecules based on favourable druglike properties, such as Absorption, Distribution, Metabolism, Excretion, and Toxicity (ADMET). We developed ADMET-AI, a machine learning platform that provides fast and accurate ADMET predictions both as a website and as a Python package. ADMET-AI has the highest average rank on the TDC ADMET Leaderboard, and it is currently the fastest web-based ADMET predictor, with a 45% reduction in time compared to the next fastest public ADMET web server. ADMET-AI can also be run locally with predictions for one million molecules taking just 3.1 h. The ADMET-AI platform is freely available both as a web server at admet.ai.greenstonebio.com and as an open-source Python package for local batch prediction at github.com/swansonk14/admet_ai (also archived on Zenodo at doi.org/10.5281/zenodo.10372930). All data and models are archived on Zenodo at doi.org/10.5281/zenodo.10372418.

A Fast and Efficient Molecular Networking Approach for Reactivity of Natural Products Exploration in Plant Extracts: Application to Diterpene Esters from Euphorbia dendroides.

Natural products represent a rich source of bioactive compounds, covering a large chemical space. Even if challenging, this diversity can be extended by applying chemical modifications. However, these studies generally require multigram amounts of isolated natural products and face frequent testing failures. To overcome this limitation, we propose a rapid and efficient approach that uses molecular networking (MN) to visualize the new chemical diversity generated by simple chemical modifications of natural extracts. Moreover, the strategy deployed enables the most appropriate reagents to be defined quickly upstream of a reaction on a pure compound, in order to maximize chemical diversity. This methodology was applied to the latex extract of Euphorbia dendroides to follow the reactivity toward a series of Brønsted and Lewis acids of three class of diterpene esters identified in this species: jatrophane, terracinolide, and phorbol. Through the molecular networking interpretation, with the aim to illustrate our approach, BF3·OEt2 was selected for chemical modification on isolated jatrophane esters. Three rearranged compounds (3-5) were obtained, showing that the most appropriate reagents can be selected by MN interpretation.

Design Principles Guided by DFT Calculations and High-Throughput Frameworks for the Discovery of New Diamond-like Chalcogenide Thermoelectric Materials.

Rational design principles are one pathway to discovering new materials. However, technological breakthroughs rarely occur in this way because these design principles are usually based on incremental advances that seldom lead to disruptive applications. The emergence of machine-learning (ML) and high-throughput (HT) techniques has changed the paradigm, opening up new possibilities for efficiently screening large chemical spaces and creating on-the-fly design principles for the discovery of novel materials with desired properties. In this work, the approach is used to discover novel thermoelectric (TE) materials based on quaternary diamond-like chalcogenides. A HT framework that integrates density functional theory calculations, ML, and the solution of the Boltzmann transport equation is used to efficiently rationalize the transport properties of these compounds and identify those with potential as TE materials, achieving ZT values above 2.

Guided Docking as a Data Generation Approach Facilitates Structure-Based Machine Learning on Kinases.

Drug discovery pipelines nowadays rely on machine learning models to explore and evaluate large chemical spaces. While including 3D structural information is considered beneficial, structural models are hindered by the availability of protein-ligand complex structures. Exemplified for kinase drug discovery, we address this issue by generating kinase-ligand complex data using template docking for the kinase compound subset of available ChEMBL assay data. To evaluate the benefit of the created complex data, we use it to train a structure-based E(3)-invariant graph neural network. Our evaluation shows that binding affinities can be predicted with significantly higher precision by models that take synthetic binding poses into account compared to ligand- or drug-target interaction models alone.

Exploring the Limits of the Generalized CHARMM and AMBER Force Fields through Predictions of Hydration Free Energy of Small Molecules.

Accurate force field parameters, potential energy functions, and receptor-ligand models are essential for modeling the solvation and binding of drug-like molecules to a receptor. A large and ever-growing chemical space of medicinally relevant scaffolds has also required these factors, especially force field parameters, to be highly transferable. Generalized force fields such as the CHARMM General Force Field (CGenFF) and the generalized AMBER force field (GAFF) have accomplished this feat along with other contemporaneous ones like OPLS. Here, we analyze the limits in the parametrization of drug-like small molecules by CGenFF and GAFF in terms of the various functional groups represented within them. Specifically, we link the presence of specific functional groups to the error in the absolute hydration free energy of over 600 small molecules, predicted by alchemical free energy methods implemented in the CHARMM program. Our investigation reveals that molecules with (i) a nitro group in CGenFF and GAFF are, respectively, over- or undersolubilized in aqueous medium, (ii) amine groups are undersolubilized more so in CGenFF than in GAFF, and (iii) carboxyl groups are more oversolubilized in GAFF than in CGenFF. We present our analyses of the potential factors underlying these trends. We also showcase the use of a machine-learning-based approach combined with the SHapley Additive exPlanations framework to attribute these trends to specific functional groups, which can be easily adopted to explore the limits of other general force fields.

SAF: Smart Aggregation Framework for Revealing Atoms Importance Rank and Improving Prediction Rates in Drug Discovery.

Machine learning, and representation learning in particular, has the potential to facilitate drug discovery by screening a large chemical space in silico. A successful approach for representing molecules is to treat them as graphs and utilize graph neural networks. One of the key limitations of such methods is the necessity to represent compounds with different numbers of atoms, which requires aggregating the atom's information. Common aggregation operators, such as averaging, result in a loss of information at the atom level. In this work, we propose a novel aggregating approach where each atom is weighted nonlinearly using the Boltzmann distribution with a hyperparameter analogous to temperature. We show that using this weighted aggregation improves the ability of the gold standard message-passing neural network to predict antibiotic activity. Moreover, by changing the temperature hyperparameter, our approach can reveal the atoms that are important for activity prediction in a smooth and consistent way, thus providing a novel regulated attention mechanism for graph neural networks. We further validate our method by showing that it recapitulates the functional group in β-lactam antibiotics. The ability of our approach to rank the atoms' importance for a desired function can be used within any graph neural network to provide interpretability of the results and predictions at the node level.

ML-Aided Computational Screening of 2D Materials for Photocatalytic Water Splitting.

The exploration of two-dimensional (2D) materials with exceptional physical and chemical properties is essential for the advancement of solar water splitting technologies. However, the discovery of 2D materials is currently heavily reliant on fragmented studies with limited opportunities for fine-tuning the chemical composition and electronic features of compounds. Starting from the V2DB digital library as a resource of 2D materials, we set up and execute a funnel approach that incorporates multiple screening steps to uncover potential candidates for photocatalytic water splitting. The initial screening step is based upon machine learning (ML) predicted properties, and subsequent steps involve first-principles modeling of increasing complexity, going from density functional theory (DFT) to hybrid-DFT to GW calculations. Ensuring that at each stage more complex calculations are only applied to the most promising candidates, our study introduces an effective screening methodology that may serve as a model for accelerating 2D materials discovery within a large chemical space. Our screening process yields a selection of 11 promising 2D photocatalysts.