Accelerating high-throughput virtual screening through molecular pool-based active learning

Structure-based virtual screening is an important tool in early stage drug discovery that scores the interactions between a target protein and candidate ligands. As virtual libraries continue to grow (in excess of 10⁸ molecules), so too do the resources necessary to conduct exhaustive virtual screening campaigns on these libraries. However, Bayesian optimization techniques, previously employed in other scientific discovery problems, can aid in their exploration: a surrogate structure–property relationship model trained on the predicted affinities of a subset of the library can be applied to the remaining library members, allowing the least promising compounds to be excluded from evaluation. In this study, we explore the application of these techniques to computational docking datasets and assess the impact of surrogate model architecture, acquisition function, and acquisition batch size on optimization performance. We observe significant reductions in computational costs; for example, using a directed-message passing neural network we can identify 94.8% or 89.3% of the top-50 000 ligands in a 100M member library after testing only 2.4% of candidate ligands using an upper confidence bound or greedy acquisition strategy, respectively. Such model-guided searches mitigate the increasing computational costs of screening increasingly large virtual libraries and can accelerate high-throughput virtual screening campaigns with applications beyond docking.

Bayesian optimization can accelerate structure-based virtual screening campaigns by minimizing the total number of simulations performed while still identifying the vast majority of computational hits.     

Automated discovery of noncovalent inhibitors of SARS-CoV-2 main protease by consensus Deep Docking of 40 billion small molecules

Recent explosive growth of ‘make-on-demand’ chemical libraries brought unprecedented opportunities but also significant challenges to the field of computer-aided drug discovery. To address this expansion of the accessible chemical universe, molecular docking needs to accurately rank billions of chemical structures, calling for the development of automated hit-selecting protocols to minimize human intervention and error. Herein, we report the development of an artificial intelligence-driven virtual screening pipeline that utilizes Deep Docking with Autodock GPU, Glide SP, FRED, ICM and QuickVina2 programs to screen 40 billion molecules against SARS-CoV-2 main protease (Mpro). This campaign returned a significant number of experimentally confirmed inhibitors of Mpro enzyme, and also enabled to benchmark the performance of twenty-eight hit-selecting strategies of various degrees of stringency and automation. These findings provide new starting scaffolds for hit-to-lead optimization campaigns against Mpro and encourage the development of fully automated end-to-end drug discovery protocols integrating machine learning and human expertise.

Deep learning-accelerated docking coupled with computational hit selection strategies enable the identification of inhibitors for the SARS-CoV-2 main protease from a chemical library of 40 billion small molecules.     

Docking rings in a solid: reversible assembling of pseudorotaxanes inside a zirconium metal–organic framework

An unprecedented zirconium metal–organic framework featuring a T-shaped benzimidazole strut was constructed and employed as a sponge-like material for selective absorption of macrocyclic guests. The neutral benzimidazole domain of the as-synthesized framework can be readily protonated and fully converted to benzimidazolium. Mechanical threading of [24]crown-8 ether wheels onto recognition sites to form pseudorotaxanes was evidenced by solution nuclear magnetic resonance, solid-state fluorescence, and infrared spectroscopy. Selective absorption of [24]crown-8 ether rather than its dibenzo counterpart was also observed. Further study reveals that this binding process is reversible and acid–base switchable. The success of docking macrocyclic guests in crystals via host–guest interactions provides an alternative route to complex functional materials with interpenetrated structures.

A T-shaped ligand was designed as struts for building a zirconium metal–organic framework. Acid–base switchable docking and releasing a 24-membered crown ether inside crystals was successfully accomplished via post-synthetic modification.     

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.     

“Click handle”-modified 2′-deoxy-2′-fluoroarabino nucleic acid as a synthetic genetic polymer capable of post-polymerization functionalization

The functions of natural nucleic acids such as DNA and RNA have transcended genetic information carriers and now encompass affinity reagents, molecular catalysts, nanostructures, data storage, and many others. However, the vulnerability of natural nucleic acids to nuclease degradation and the lack of chemical functionality have imposed a significant constraint on their ever-expanding applications. Herein, we report the synthesis and polymerase recognition of a 5-(octa-1,7-diynyl)uracil 2′-deoxy-2′-fluoroarabinonucleic acid (FANA) triphosphate. The DNA-templated, polymerase-mediated primer extension using this “click handle”-modified FANA (cmFANA) triphosphate and other FANA nucleotide triphosphates consisting of canonical nucleobases efficiently generated full-length products. The resulting cmFANA polymers exhibited excellent nuclease resistance and the ability to undergo efficient click conjugation with azide-functionalized molecules, thereby becoming a promising platform for serving as a programmable and evolvable synthetic genetic polymer capable of post-polymerization functionalization.

Polymerase-mediated incorporation of a “click handle”-modified fluoroarabinonucleic acid (cmFANA) triphosphate produces a new class of nuclease-resistant, evolvable genetic polymers that can be functionalized with azide-containing molecules.     

An endogenous stimulus detonated nanocluster-bomb for contrast-enhanced cancer imaging and combination therapy

Exploitation of stimuli-responsive nanoplatforms is of great value for precise and efficient cancer theranostics. Herein, an in situ activable “nanocluster-bomb” detonated by endogenous overexpressing legumain is fabricated for contrast-enhanced tumor imaging and controlled gene/drug release. By utilizing the functional peptides as bioligands, TAMRA-encircled gold nanoclusters (AuNCs) endowed with targeting, positively charged and legumain-specific domains are prepared as quenched building blocks due to the AuNCs'' nanosurface energy transfer (NSET) effect on TAMRA. Importantly, the AuNCs can shelter therapeutic cargos of DNAzyme and Dox (Dzs-Dox) to aggregate larger nanoparticles as a “nanocluster-bomb” (AuNCs/Dzs-Dox), which could be selectively internalized into cancer cells by integrin-mediated endocytosis and in turn locally hydrolyzed in the lysosome with the aid of legumain. A “bomb-like” behavior including “spark-like” appearance (fluorescence on) derived from the diminished NSET effect of AuNCs and cargo release (disaggregation) of Dzs-Dox is subsequently monitored. The results showed that the AuNC-based disaggregation manner of the “nanobomb” triggered by legumain significantly improved the imaging contrast due to the activable mechanism and the enhanced cellular uptake of AuNCs. Meanwhile, the in vitro cytotoxicity tests revealed that the detonation strategy based on AuNCs/Dzs-Dox readily achieved efficient gene/chemo combination therapy. Moreover, the super efficacy of combinational therapy was further demonstrated by treating a xenografted MDA-MB-231 tumor model in vivo. We envision that our multipronged design of theranostic “nanocluster-bomb” with endogenous stimuli-responsiveness provides a novel strategy and great promise in the application of high contrast imaging and on-demand drug delivery for precise cancer theranostics.

An in situ activable “nanocluster-bomb” detonated by endogenous overexpressing legumain is fabricated for contrast enhanced cancer imaging and effective gene/chemo-therapy.     

Density functionals with asymptotic-potential corrections are required for the simulation of spectroscopic properties of materials

Five effects of correction of the asymptotic potential error in density functionals are identified that significantly improve calculated properties of molecular excited states involving charge-transfer character. Newly developed materials-science computational methods are used to demonstrate how these effects manifest in materials spectroscopy. Connection is made considering chlorophyll-a as a paradigm for molecular spectroscopy, 22 iconic materials as paradigms for 3D materials spectroscopy, and the V_N⁻ defect in hexagonal boron nitride as an example of the spectroscopy of defects in 2D materials pertaining to nanophotonics. Defects can equally be thought of as being “molecular” and “materials” in nature and hence bridge the relms of molecular and materials spectroscopies. It is concluded that the density functional HSE06, currently considered as the standard for accurate calculations of materials spectroscopy, should be replaced, in most instances, by the computationally similar but asymptotically corrected CAM-B3LYP functional, with some specific functionals for materials-use only providing further improvements.

Spectroscopic transitions in materials that involve charge transfer require asymptotically corrected density functionals. As most transitions do have some charge transfer character, use of such methods are generally warranted.     

The (not so) simple prediction of enantioselectivity – a pipeline for high-fidelity computations

The computation of reaction selectivity represents an appealing complementary route to experimental studies and a powerful means to refine catalyst design strategies. Accurately establishing the selectivity of reactions facilitated by molecular catalysts, however, remains a challenging task for computational chemistry. The small free energy differences that lead to large variations in the enantiomeric ratio (er) represent particularly tricky quantities to predict with sufficient accuracy to be helpful for prioritizing experiments. Further complicating this problem is the fact that standard approaches typically consider only one or a handful of conformers identified through human intuition as pars pro toto of the conformational space. Obviously, this assumption can potentially lead to dramatic failures should key energetic low-lying structures be missed. Here, we introduce a multi-level computational pipeline leveraging the graph-based Molassembler library to construct an ensemble of molecular catalysts. The manipulation and interpretation of molecules as graphs provides a powerful and direct route to tailored functionalization and conformer generation that facilitates high-throughput mechanistic investigations of chemical reactions. The capabilities of this approach are validated by examining a Rh(iii) catalyzed asymmetric C–H activation reaction and assessing the limitations associated with the underlying ligand design model. Specifically, the presence of remarkably flexible chiral Cp ligands, which induce the experimentally observed high level of selectivity, present a rich configurational landscape where multiple unexpected conformations contribute to the reported enantiomeric ratios (er). Using Molassembler, we show that considering about 20 transition state conformations per catalysts, which are generated with little human intervention and are not tied to “back-of-the-envelope” models, accurately reproduces experimental er values with limited computational expense.

The computation of reaction selectivity represents an appealing complementary route to experimental studies and a powerful mean to refine catalyst design strategies.     

Understanding the geometric diversity of inorganic and hybrid frameworks through structural coarse-graining

Much of our understanding of complex structures is based on simplification: for example, metal–organic frameworks are often discussed in the context of “nodes” and “linkers”, allowing for a qualitative comparison with simpler inorganic structures. Here we show how such an understanding can be obtained in a systematic and quantitative framework, combining atom-density based similarity (kernel) functions and unsupervised machine learning with the long-standing idea of “coarse-graining” atomic structure. We demonstrate how the latter enables a comparison of vastly different chemical systems, and we use it to create a unified, two-dimensional structure map of experimentally known tetrahedral AB₂ networks – including clathrate hydrates, zeolitic imidazolate frameworks (ZIFs), and diverse inorganic phases. The structural relationships that emerge can then be linked to microscopic properties of interest, which we exemplify for structural heterogeneity and tetrahedral density.

A coarse-graining approach enables structural comparisons across vastly different chemical spaces, from inorganic polymorphs to hybrid framework materials.     

Natural product drug discovery in the artificial intelligence era

Natural products (NPs) are primarily recognized as privileged structures to interact with protein drug targets. Their unique characteristics and structural diversity continue to marvel scientists for developing NP-inspired medicines, even though the pharmaceutical industry has largely given up. High-performance computer hardware, extensive storage, accessible software and affordable online education have democratized the use of artificial intelligence (AI) in many sectors and research areas. The last decades have introduced natural language processing and machine learning algorithms, two subfields of AI, to tackle NP drug discovery challenges and open up opportunities. In this article, we review and discuss the rational applications of AI approaches developed to assist in discovering bioactive NPs and capturing the molecular “patterns” of these privileged structures for combinatorial design or target selectivity.

Natural products (NPs) are primarily recognized as privileged structures to interact with protein drug targets.     

In-solution direct oxidative coupling for the integration of sulfur/selenium into DNA-encoded chemical libraries

Sulfur/selenium-containing electron-rich arenes (ERAs) exist in a wide range of both approved and investigational drugs with diverse pharmacological activities. These unique chemical structures and bioactive properties, if combined with the emerging DNA-encoded chemical library (DEL) technique, would facilitate drug and chemical probe discovery. However, it remains challenging, as there is no general DNA-compatible synthetic methodology available for the formation of C–S and C–Se bonds in aqueous solution. Herein, an in-solution direct oxidative coupling procedure that could efficiently integrate sulfur/selenium into the ERA under mild conditions is presented. This method features simple DNA-conjugated electron-rich arenes with a broad substrate scope and a transition-metal free process. Furthermore, this synthetic methodology, examined by a scale-up reaction test and late-stage precise modification in a mock peptide-like DEL synthesis, will enable its utility for the synthesis of sulfur/selenium-containing DNA-encoded libraries and the discovery of bioactive agents.

DNA-compatible direct oxidative coupling using various sulfur/selenium sources has been achieved, featuring pre-functionalization-free substrates and transition metal-free condition.     

Zn(ii)-Coordination-driven self-assembled nanoagents for multimodal imaging-guided photothermal/gene synergistic therapy

Synergistic photothermal therapy (PTT) with gene therapy (GT) has drawn emerging interest in the improvement of cancer therapeutic efficiency, while the co-delivery of photothermal agents (PTAs) and therapeutic genes by an integrated nanoplatform, with controllability and biodegradability, is still challenging and urgently desired. Herein, a multi-functional metal–organic framework (MOF) based PTT–GT platform (siRNA@PT-ZIF-8) was developed, which was constructed with siRNA, a near-infrared (NIR) responsive organic dye IR780 derivative (IR780-1), and 2-methylimidazole (2-MIM) by a facile one-pot self-assembly method. This “all-in-one” system of siRNA@PT-ZIF-8 enabled not only photothermal/photoacoustic/fluorescence multimodal imaging but also tumor microenvironment responsiveness for specific and on-demand release of therapeutic cargos, overcoming the inherent limitations of free gene or organic PTA molecules (e.g., short blood circulation half-life and weak stability) in conventional PTT and GT. This nanoplatform provides an efficient and safe strategy for cancer theranostics, and the one-step assembly strategy favors personalized formulation design for diverse demands in cancer management.

siRNA@PT-ZIF-8 was prepared by one pot self-assembly for tri-mode imaging guided mild-temperature photothermal synergetic gene therapy.     

SiRNA-templated 3D framework nucleic acids for chemotactic recognition,and programmable and visualized precise delivery for synergistic cancer therapy

Developments in framework nucleic acids (FNAs) are limited by complicated synthesis, by-product interference, and low framework utilization. Herein, simple core–shell spherical 3D FNAs (ST-SFNAs) preparation is presented based on siRNA-templated linear polymerization followed by hybridization chain reaction branched polymerization. Without by-products, all components exhibited their special functions to obtain high space utilization of ST-SFNAs. ST-SFNAs were covered by catalase and folic acid-functionalized liposome membranes. The catalase endowed ST-SFNAs with chemotactic activities in the H₂O₂ reaction catalyzed by catalase. Furthermore, combined with functionalized folic acids'' targeting folate receptors, the synergistic chemotactic recognition of cancer cells was obtained. This dramatically promoted targeted cellular uptakes compared with traditional active or passive targeting pathways. Subsequently, the cascaded-logical programmable release of drugs was precisely controlled by targeting glutathione and ATP (via S–S bond and ATP aptamer on the inner g-DNA cover). This was visualized by “turn on” fluorescent signals generated by special hybridization of released hairpin DNAs with survivin mRNA biomarkers. Simultaneously, biocompatible synergistic therapy was achieved by simultaneously releasing doxorubicin and siRNA. With its high utilization for synergistic chemotactic recognition, programmable and visualized delivery, as well as synergistic therapy, an efficient platform for maximizing the therapeutic efficacy has been developed. This would initiate further FNA-based material development for a variety of biological applications.

ST-SFNAs were developed by siRNA-templated linear-branched polymerizations with high space utilizations for loadings, which obtained synergistic cancer therapy via chemotactic recognition, visualized delivery, and cascaded-logical controlled release.     

Computational and experimental investigation of the effect of cation structure on the solubility of anionic flow battery active-materials

Recent advances in clean, sustainable energy sources such as wind and solar have enabled significant cost improvements, yet their inherent intermittency remains a considerable challenge for year-round reliability demanding the need for grid-scale energy storage. Nonaqueous redox flow batteries (NRFBs) have the potential to address this need, with attractive attributes such as flexibility to accommodate long- and short-duration storage, separately scalable energy and power ratings, and improved safety profile over integrated systems such as lithium-ion batteries. Currently, the low-solubility of NRFB electrolytes fundamentally limits their energy density. However, synthetically exploring the large chemical and parameter space of NRFB active materials is not only costly but also intractable. Here, we report a computational framework, coupled with experimental validation, designed to predict the solubility trends of electrolytes, incorporating both the lattice and solvation free energies. We reveal that lattice free energy, which has previously been neglected, has a significant role in tuning electrolyte solubility, and that solvation free energies alone is insufficient. The desymmetrization of the alkylammonium cation leading to short-chain, asymmetric cations demonstrated a modest increase in solubility, which can be further explored for NRFB electrolyte development and optimization. The resulting synergistic computational–experimental approach provides a cost-effective strategy in the development of high-solubility active materials for high energy density NRFB systems.

Active-material solubility is critical in determining NRFB energy density, yet a predictive model accounting for solid-state cohesion energy has remained elusive. Herein we present such, based on an empirically calibrated computational framework.     

Covalent flexible peptide docking in Rosetta

Electrophilic peptides that form an irreversible covalent bond with their target have great potential for binding targets that have been previously considered undruggable. However, the discovery of such peptides remains a challenge. Here, we present Rosetta CovPepDock, a computational pipeline for peptide docking that incorporates covalent binding between the peptide and a receptor cysteine. We applied CovPepDock retrospectively to a dataset of 115 disulfide-bound peptides and a dataset of 54 electrophilic peptides. It produced a top-five scoring, near-native model, in 89% and 100% of the cases when docking from the native conformation, and 20% and 90% when docking from an extended peptide conformation, respectively. In addition, we developed a protocol for designing electrophilic peptide binders based on known non-covalent binders or protein–protein interfaces. We identified 7154 peptide candidates in the PDB for application of this protocol. As a proof-of-concept we validated the protocol on the non-covalent complex of 14-3-3σ and YAP1 phosphopeptide. The protocol identified seven highly potent and selective irreversible peptide binders. The predicted binding mode of one of the peptides was validated using X-ray crystallography. This case-study demonstrates the utility and impact of CovPepDock. It suggests that many new electrophilic peptide binders can be rapidly discovered, with significant potential as therapeutic molecules and chemical probes.

We developed Rosetta CovPepDock, a computational pipeline for covalent peptide docking. We showed it is highly accurate in retrospective benchmarks, and applied it prospectively to design potent and selective covalent binders of 14-3-3σ.     

Optical resolution of 1,16-dihydroxytetraphenylene by chiral gold(iii) complexation and its applications as chiral ligands in asymmetric catalysis

We report herein a novel approach involving optical resolution of (±)-1,16-dihydroxytetraphenylene (DHTP) by chiral gold(iii) complexation. This method features several key advantages, i.e., recyclability of chiral resolution reagents, feasibility of scaling up to gram quantities, and operational simplicity. On the basis of this method, which led to optically pure DHTP, a library of 2,15-diaryl (S)-DHTPs and several (S)-DHTP-derived phosphoramidite ligands were synthesized. Finally, the superior performance of a (S)-DHTP phosphoramidite ligand was demonstrated by efficient iridium-catalyzed asymmetric allylic alkynylation reactions.

Chiral gold(iii) complex was found to be a “golden key” to unlock (±)-1,16-dihydroxytetraphenylene (DHTP), leading to enantiopure DHTP in a large quantity. This approach provided convenient access to 2,15-diaryl (S)-DHTPs and phosphoramidites.     

Catalytic oxidative dehydrogenation of N-heterocycles with nitrogen/phosphorus co-doped porous carbon materials

A metal-free oxidative dehydrogenation of N-heterocycles utilizing a nitrogen/phosphorus co-doped porous carbon (NPCH) catalyst is reported. The optimal material is robust against traditional poisoning agents and shows high antioxidant resistance. It exhibits good catalytic performance for the synthesis of various quinoline, indole, isoquinoline, and quinoxaline ‘on-water’ under air atmosphere. The active sites in the NPCH catalyst are proposed to be phosphorus and nitrogen centers within the porous carbon network.

Green oxidations made easy. Metal-free dehydrogenation of N-heterocycles are possible in using N,P-co-doped porous carbon materials “on” water using air.     

Palladium-catalyzed asymmetric allylic alkylation (AAA) with alkyl sulfones as nucleophiles

An efficient palladium-catalyzed AAA reaction with a simple α-sulfonyl carbon anion as nucleophiles is presented for the first time. Allyl fluorides are used as superior precursors for the generation of π-allyl complexes that upon ionization liberate fluoride anions for activation of silylated nucleophiles. With the unique bidentate diamidophosphite ligand ligated palladium as catalyst, the in situ generated α-sulfonyl carbon anion was quickly captured by the allylic intermediates, affording a series of chiral homo-allylic sulfones with high efficiency and selectivity. This work provides a mild in situ desilylation strategy to reveal nucleophilic carbon centers that could be used to overcome the pK_a limitation of “hard” nucleophiles in enantioselective transformations.

A variety of “hard” α-sulfonyl carbanions of aryl, heteroaryl and alkyl sulfones were successfully employed as nucleophiles in palladium-catalyzed asymmetric allylic alkylation with excellent enantioselectivities.