Exploring Chemical Reaction Space with Machine Learning Models: Representation and Feature Perspective.

Chemical reactions serve as foundational building blocks for organic chemistry and drug design. In the era of large AI models, data-driven approaches have emerged to innovate the design of novel reactions, optimize existing ones for higher yields, and discover new pathways for synthesizing chemical structures comprehensively. To effectively address these challenges with machine learning models, it is imperative to derive robust and informative representations or engage in feature engineering using extensive data sets of reactions. This work aims to provide a comprehensive review of established reaction featurization approaches, offering insights into the selection of representations and the design of features for a wide array of tasks. The advantages and limitations of employing SMILES, molecular fingerprints, molecular graphs, and physics-based properties are meticulously elaborated. Solutions to bridge the gap between different representations will also be critically evaluated. Additionally, we introduce a new frontier in chemical reaction pretraining, holding promise as an innovative yet unexplored avenue.

CMNPD: a comprehensive marine natural products database towards facilitating drug discovery from the ocean.

Marine organisms are expected to be an important source of inspiration for drug discovery after terrestrial plants and microorganisms. Despite the remarkable progress in the field of marine natural products (MNPs) chemistry, there are only a few open access databases dedicated to MNPs research. To meet the growing demand for mining and sharing for MNPs-related data resources, we developed CMNPD, a comprehensive marine natural products database based on manually curated data. CMNPD currently contains more than 31 000 chemical entities with various physicochemical and pharmacokinetic properties, standardized biological activity data, systematic taxonomy and geographical distribution of source organisms, and detailed literature citations. It is an integrated platform for structure dereplication (assessment of novelty) of (marine) natural products, discovery of lead compounds, data mining of structure-activity relationships and investigation of chemical ecology. Access is available through a user-friendly web interface at https://www.cmnpd.org. We are committed to providing a free data sharing platform for not only professional MNPs researchers but also the broader scientific community to facilitate drug discovery from the ocean.

Chemical Space, Scaffolds, and Halogenated Compounds of CMNPD: A Comprehensive Chemoinformatic Analysis.

The comprehensive marine natural products database (CMNPD) is a new free access and comprehensive database developed originally by Lyu's team of our research group, including more than 30 000 marine natural products (MNPs) reported from the 1960s. In this article, we aimed to present CMNPD's value in drug discovery and to present several characteristics of MNPs based on our new comprehensive data. We used chemoinformatic analysis methods to report the molecular properties, chemical space, and several scaffold assessments of CMNPD compared with several databases. Then, we reported the characteristics of MNPs from the aspect of halogens, comparing MNPs with terrestrial natural products (TNPs) and drugs. We found that CMNPD had a low proportion (2.91%) of scaffolds utilized by drugs, and high similarities between CMNPD and NPAtlas (a microbial natural products database), which are worth further investigation. The proportion of bromides in MNPs is outstandingly higher (11.0%) in contrast to other halogens. Furthermore, the results showed great differences in halogenated structures between MNPs and drugs, especially brominated substructures. Finally, we found that many marine species (2.52%) reported only halogenated compounds. It can be concluded from these results that CMNPD is a promising source for drug discovery and has many scientific issues relative to MNPs that need to be further investigated.

Target Prediction Model for Natural Products Using Transfer Learning.

A large proportion of lead compounds are derived from natural products. However, most natural products have not been fully tested for their targets. To help resolve this problem, a model using transfer learning was built to predict targets for natural products. The model was pre-trained on a processed ChEMBL dataset and then fine-tuned on a natural product dataset. Benefitting from transfer learning and the data balancing technique, the model achieved a highly promising area under the receiver operating characteristic curve (AUROC) score of 0.910, with limited task-related training samples. Since the embedding distribution difference is reduced, embedding space analysis demonstrates that the model's outputs of natural products are reliable. Case studies have proved our model's performance in drug datasets. The fine-tuned model can successfully output all the targets of 62 drugs. Compared with a previous study, our model achieved better results in terms of both AUROC validation and its success rate for obtaining active targets among the top ones. The target prediction model using transfer learning can be applied in the field of natural product-based drug discovery and has the potential to find more lead compounds or to assist researchers in drug repurposing.

Functional Characterization and Structural Basis of an Efficient Di-C-glycosyltransferase from Glycyrrhiza glabra.

A highly efficient di-C-glycosyltransferase GgCGT was discovered from the medicinal plant Glycyrrhiza glabra. GgCGT catalyzes a two-step di-C-glycosylation of flopropione-containing substrates with conversion rates of >98%. To elucidate the catalytic mechanisms of GgCGT, we solved its crystal structures in complex with UDP-Glc, UDP-Gal, UDP/phloretin, and UDP/nothofagin, respectively. Structural analysis revealed that the sugar donor selectivity was controlled by the hydrogen-bond interactions of sugar hydroxyl groups with D390 and other key residues. The di-C-glycosylation capability of GgCGT was attributed to a spacious substrate-binding tunnel, and the G389K mutation could switch di- to mono-C-glycosylation. GgCGT is the first di-C-glycosyltransferase with a crystal structure, and the first C-glycosyltransferase with a complex structure containing a sugar acceptor. This work could benefit the development of efficient biocatalysts to synthesize C-glycosides with medicinal potential.

Engineering a bacterial sialyltransferase for di-sialylation of a therapeutic antibody.

Terminal α-2,6-sialylation of N-glycans is a humanized glycosylation that affects the properties and efficacy of therapeutic glycoproteins. Fc di-sialylation (a biantennary N-glycan with two α-2,6-linked sialic acids) of IgG antibodies imparts them with enhanced anti-inflammatory activity and other roles. However, the microheterogeneity of N-glycoforms presents a challenge for therapeutic development. Therefore, controlled sialylation has drawn considerable attention, but direct access to well-defined di-sialylated antibodies remains limited. Herein, a one-pot three-enzyme protocol was developed by engineering a bacterial sialyltransferase to facilitate the modification of therapeutic antibodies with N-acetylneuraminic acid or its derivatives towards optimized glycosylation. To overcome the low proficiency of bacterial sialyltransferase in antibody remodeling, the Photobacterium sp. JT-ISH-224 α-2,6-sialyltransferase (Psp2,6ST) was genetically engineered by terminal truncation and site-directed mutagenesis based on its protein crystal structure. With the optimized reaction conditions and using activity-based screening of various Psp2,6ST variants, a truncated mutant Psp2,6ST (111-511)-His6 A235M/A366G was shown to effectively improve the catalytic efficiency of antibody di-sialylation. Herceptin and the donor substrate promiscuity allow the introduction of bioorthogonal modifications of N-acetylneuraminic acid into antibodies for site-specific conjugation. 2-AB hydrophilic interaction chromatography analysis of the released N-glycans and intact mass characterization confirmed the high di-sialylation of Herceptin via the optimized one-pot three-enzyme reaction. This study established a versatile enzymatic approach for producing highly di-sialylated IgG antibodies. It provides new insights into engineering bacterial sialyltransferase for adaptation to the enzymatic glycoengineering of therapeutic antibodies and the glycosite-specific conjugation of antibodies.

DeepScaffold: A Comprehensive Tool for Scaffold-Based De Novo Drug Discovery Using Deep Learning.

The ultimate goal of drug design is to find novel compounds with desirable pharmacological properties. Designing molecules retaining particular scaffolds as their core structures is an efficient way to obtain potential drug candidates. We propose a scaffold-based molecular generative model for drug discovery, which performs molecule generation based on a wide spectrum of scaffold definitions, including Bemis-Murcko scaffolds, cyclic skeletons, and scaffolds with specifications on side-chain properties. The model can generalize the learned chemical rules of adding atoms and bonds to a given scaffold. The generated compounds were evaluated by molecular docking in DRD2 targets, and the results demonstrated that this approach can be effectively applied to solve several drug design problems, including the generation of compounds containing a given scaffold and de novo drug design of potential drug candidates with specific docking scores.

TF3P: Three-Dimensional Force Fields Fingerprint Learned by Deep Capsular Network.

Molecular fingerprints are the workhorse in ligand-based drug discovery. In recent years, an increasing number of research papers reported fascinating results on using deep neural networks to learn 2D molecular representations as fingerprints. It is anticipated that the integration of deep learning would also contribute to the prosperity of 3D fingerprints. Here, we unprecedentedly introduce deep learning into 3D small molecule fingerprints, presenting a new one we termed as the three-dimensional force fields fingerprint (TF3P). TF3P is learned by a deep capsular network whose training is in no need of labeled data sets for specific predictive tasks. TF3P can encode the 3D force fields information of molecules and demonstrates the stronger ability to capture 3D structural changes, to recognize molecules alike in 3D but not in 2D, and to identify similar targets inaccessible by other 2D or 3D fingerprints based on only ligands similarity. Furthermore, TF3P is compatible with both statistical models (e.g., similarity ensemble approach) and machine learning models. Altogether, we report TF3P as a new 3D small molecule fingerprint with a promising future in ligand-based drug discovery. All codes are written in Python and available at https://github.com/canisw/tf3p.

Pose Filter-Based Ensemble Learning Enables Discovery of Orally Active, Nonsteroidal Farnesoid X Receptor Agonists.

Farnesoid X receptor (FXR) agonists can reverse dysregulated bile acid metabolism, and thus, they are potential therapeutics to prevent and treat nonalcoholic fatty liver disease. The low success rate of FXR agonists' R&D and the side effects of clinical candidates such as obeticholic acid make it urgent to discover new chemotypes. Unfortunately, structure-based virtual screening (SBVS) that can speed up drug discovery has rarely been reported with success for FXR, which was likely hindered by the failure in addressing protein flexibility. To address this issue, we devised human FXR (hFXR)-specific ensemble learning models based on pose filters from 24 agonist-bound hFXR crystal structures and coupled them to traditional SBVS approaches of the FRED docking plus Chemgauss4 scoring function. It turned out that the hFXR-specific pose filter ensemble (PFE) was able to improve ligand enrichment significantly, which rendered 3RUT-based SBVS with its PFE the ideal approach for FXR agonist discovery. By screening of the Specs chemical library and in vitro FXR transactivation bioassay, we identified a new class of FXR agonists with compound XJ034 as the representative, which would have been missed if the PFE was not coupled. Following that, we performed in-depth biological studies which demonstrated that XJ034 resulted in a downtrend of intracellular triglyceride in vitro, significantly decreased the serum/liver TG in high fat diet-induced C57BL/6J obese mice, and more importantly, showed metabolic stabilities in both plasma and liver microsomes. To provide insight into further structure-based lead optimization, we solved the crystal structure of hFXR complexed with compound XJ034, uncovering a unique hydrogen bond between compound XJ034 and residue Y375. The current work highlights the power of our pose filter-based ensemble learning approach in terms of scaffold hopping and provides a promising lead compound for further development.

Discovery and antibacterial study of potential PPK1 inhibitors against uropathogenic E. coli.

Novel antibacterial agents are urgently needed to address the infections caused by multi-drug resistant bacteria. Urinary tract infections are common infectious diseases in clinical. Most of these infections are caused by drug-resistant uropathogenic Escherichia coli. PPK1 is an essential kinase for bacterial motility, biofilm formation, quorum sensing, and virulence factors in the expression of uropathogenic E. coli. In the present study, two small molecules potentially targeting PPK1 were discovered through virtual screening and biological assays. The in vitro and in vivo results suggested that the interaction of these compounds with PPK1 can disrupt biofilm formation of uropathogenic E. coli and reduce invasive ability and resistance to oxidative stress of this strain. Moreover, the compounds exhibit good antibacterial bacterial activity in the mice with urinary tract infection. Taken together, our findings could provide a new chemotype for the development of antibacterials targeting PPK1.

Rational Design and Identification of Small-Molecule Allosteric Inhibitors of CD38.

CD38 is a multi-functional signaling enzyme that catalyzes the biosynthesis of two calcium-mobilizing second messengers: cyclic ADP-ribose and nicotinic acid adenine dinucleotide phosphate. It also regulates intracellular nicotinamide adenine dinucleotide (NAD) contents, associated with multiple pathophysiological processes such as aging and cancer. As such, enzymatic inhibitors of CD38 offer great potential in drug development. Here, through virtual screening and enzymatic assays, we discovered compound LX-102, which targets CD38 on the side opposite its enzymatic pocket with a binding affinity of 7.7 µm. It inhibits the NADase activity of CD38 with an IC50 of 14.9 µm. Surface plasmon resonance (SPR) and hydrogen/deuterium exchange and mass spectrometry experiments verified that LX-102 competitively binds to the epitope of the therapeutic SAR 650984 antibody in an allosteric manner. Molecular dynamics simulation was performed to demonstrate the binding dynamics of CD38 with the allosteric ligand. In summary, we established that the cavity to which SAR 650984 binds was an allosteric site and was accessible for the rational design of small chemical modulators of CD38. The lead compound LX-102 that we identified in this study could also be a useful tool for probing CD38 functions and promoting drug discovery.

Graph-based generative models for de Novo drug design.

The discovery of new chemical entities is a crucial part of drug discovery, which requires the lead compounds to have desired properties to be pharmaceutically active. De novo drug design aims to generate and optimize novel ligands for macromolecular targets from scratch. The development of graph-based deep generative neural networks has provided a new method. In this review, we gave a brief introduction to graph representation and graph-based generative models for de novo drug design, summarized them as four architectures, and concluded each's characteristics. We also discussed generative models for scaffold- and fragment-based design and graph-based generative models' future directions.

Calcium-Mobilizing Behaviors of Neutral Cyclic ADP-Ribose Mimics that Integrate Modifications to the Nucleobase, Northern Ribose and Pyrophosphate.

Cyclic adenosine diphosphate ribose (cADPR) is an endogenous Ca2+ mobilizer involved in diverse cellular processes. Mimics of cADPR play a crucial role in investigating the molecular mechanism(s) of cADPR-mediated signaling. Here, compound 3, a mimic of cADPR in which a neutral triazole moiety and an ether linkage were introduced to substitute the pyrophosphate and "northern" ribose components, respectively, was synthesized for the first time. The pharmacological activities in Jurkat cells indicated that this mimic is capable of penetrating plasma membrane and inciting Ca2+ release from the endoplasmic reticulum (ER) through the action of ryanodine receptors (RyRs) and triggering Ca2+ influx. Furthermore, a uridine moiety was introduced in place of adenine and the new cADPR mimics 4 and 5 were synthesized. The results of biological investigation showed that these mimics also targeted RyRs and retained moderate Ca2+ agonistic activities. The results indicated that the neutral cADPR mimics had the same targets for inducing Ca2+ signaling.

Discovery of new GSK-3ß inhibitors through structure-based virtual screening.

Glycogen synthase kinase-3ß (GSK-3ß) is an attractive therapeutic target for human diseases, such as diabetes, cancer, neurodegenerative diseases, and inflammation. Thus, structure-based virtual screening was performed to identify novel scaffolds of GSK-3ß inhibitors, and we observed that conserved water molecules of GSK-3ß were suitable for virtual screening. We found 14 hits and D1 (IC50 of 0.71 µM) were identified. Furthermore, the neuroprotection activity of D1-D3 was validated on a cellular level. 2D similarity searches were used to find derivatives of high inhibitory compounds and an enriched structure-activity relationship suggested that these skeletons were worthy of study as potent GSK-3ß inhibitors.

Maximal Unbiased Benchmarking Data Sets for Human Chemokine Receptors and Comparative Analysis.

Chemokine receptors (CRs) have long been druggable targets for the treatment of inflammatory diseases and HIV-1 infection. As a powerful technique, virtual screening (VS) has been widely applied to identifying small molecule leads for modern drug targets including CRs. For rational selection of a wide variety of VS approaches, ligand enrichment assessment based on a benchmarking data set has become an indispensable practice. However, the lack of versatile benchmarking sets for the whole CRs family that are able to unbiasedly evaluate every single approach including both structure- and ligand-based VS somewhat hinders modern drug discovery efforts. To address this issue, we constructed Maximal Unbiased Benchmarking Data sets for human Chemokine Receptors (MUBD-hCRs) using our recently developed tools of MUBD-DecoyMaker. The MUBD-hCRs encompasses 13 subtypes out of 20 chemokine receptors, composed of 404 ligands and 15756 decoys so far and is readily expandable in the future. It had been thoroughly validated that MUBD-hCRs ligands are chemically diverse while its decoys are maximal unbiased in terms of "artificial enrichment", "analogue bias". In addition, we studied the performance of MUBD-hCRs, in particular CXCR4 and CCR5 data sets, in ligand enrichment assessments of both structure- and ligand-based VS approaches in comparison with other benchmarking data sets available in the public domain and demonstrated that MUBD-hCRs is very capable of designating the optimal VS approach. MUBD-hCRs is a unique and maximal unbiased benchmarking set that covers major CRs subtypes so far.

Analysis of LncRNA expression in cell differentiation.

Lineage-specific cell differentiation is a precise and coordinated biological process. To explore the roles of long noncoding RNA (lncRNA) in this process, the expression of polyA-minus RNAs was comparatively studied during the course of myocyte and adipocyte differentiation. In addition to identifying thousands of novel lncRNAs, distinct lncRNA profiles were revealed during lineage-specific differentiation, showing their active involvement in this process. This study further found that lncRNAs were organized in clusters and are co-regulated, constituting transcription open domains (TODs). In myocyte differentiation of C2C12 cells, loss-of-function screening identified three myogenic lncRNAs. Knockdown of their expression compromised not only the differentiation process, but also the essential signaling pathway. In addition to showing that lncRNAs are actively involved in cell differentiation, our results start to reveal a comprehensive signaling pathway, involving both protein and lncRNA factors.

Design, synthesis and structure-activity relationship study of novel naphthoindolizine and indolizinoquinoline-5,12-dione derivatives as IDO1 inhibitors.

Indoleamine 2,3-dioxygenase 1 (IDO1) is regarded as a promising target for cancer immunotherapy. Many naphthoquinone derivatives have been reported as IDO1 inhibitors so far. Herein, two series of naphthoquinone derivatives, naphthoindolizine and indolizinoquinoline-5,12-dione derivatives, were synthesized and evaluated for their IDO1 inhibitory activity. Most of the target compounds showed significant inhibition potency and high selectivity for IDO1 over tryptophan 2,3-dioxygenase (TDO). The structure-activity relationship was also summarized. The most potent compounds 5c (IC50 23 nM, IDO1 enzyme), and 5b' (IC50 372 nM, HeLa cell) were identified as promising lead compounds.

Discovery of N-(Naphtho[1,2-b]Furan-5-Yl) Benzenesulfonamides as Novel Selective Inhibitors of Triple-Negative Breast Cancer (TNBC).

Any type of breast cancer not expressing genes of the estrogen receptor (ER), progesterone receptor (PR), or human epidermal growth factor receptor 2 (HER2) is referred to as triple-negative breast cancer (TNBC). Accordingly, TNBCs do not respond to hormonal therapies or medicines targeting the ER, PR, or HER2. Systemic chemotherapy is therefore the only treatment option available today and prognoses remain poor. We report the discovery and characterization of N-(naphtho[1,2-b]furan-5-yl)benzenesulfonamides as selective inhibitors of TNBCs. These inhibitors were identified by virtual screening and inhibited different TNBC cell lines with IC50 values of 2-3 µM. The compounds did not inhibit normal (i.e. MCF-7 and MCF-10A) cells in vitro, indicating their selectivity against TNBC cells. Considering the selectivity of these inhibitors for TNBC, these compounds and analogs can serve as a promising starting point for further research on effective TNBC inhibitors.

Scale-Up Synthesis and Identification of GLYX-13, a NMDAR Glycine-Site Partial Agonist for the Treatment of Major Depressive Disorder.

GLYX-13, a NMDAR glycine-site partial agonist, was discovered as a promising antidepressant with rapidly acting effects but no ketamine-like side effects. However, the reported synthetic process route had deficiencies of low yield and the use of unfriendly reagents. Here, we report a scaled-up synthesis of GLYX-13 with an overall yield of 30% on the hectogram scale with a column chromatography-free strategy, where the coupling and deprotection reaction conditions were systematically optimized. Meanwhile, the absolute configuration of precursor compound of GLYX-13 was identified by X-ray single crystal diffraction. Finally, the activity of GLYX-13 was verified in the cortical neurons of mice through whole-cell voltage-clamp technique.

[Analysis of traditional Chinese medicine prescriptions based on support vector machine and analytic hierarchy process].

Combined use of drugs is a hot spot in the research of new drugs nowadays, and traditional Chinese medicine (TCM) is a classic practice in the combined use of drugs. In this paper, the compatibility of TCM prescriptions and the related properties of composed herbs were calculated and studied to verify and discuss the feasibility of the results in guiding compatibility. Research Group on New Drug Design, Institute of Chinese Materia Medica, China Academy of Chinese Medical Sciences had established a structured database of TCM prescriptions by using traditional Chinese medicine inheritance support system (TCMISS V2.0), including 4 012 prescription compatibilities, 2 072 drug components, 381 kinds of TCM diseases, 316 kinds of TCM syndromes and 26 kinds of drug properties. On the basis of the created database above, Support Vector Machine (SVM) was used to analyze the prescription compatibility data and establish a model for predicting feasibility of drug compatibilities. Analytic Hierarchy Process (AHP) and cluster analysis were used to study the influence of drug properties in the rationality of prescription compatibility. The computational results showed that the accuracy in efficacy prediction of two data sets, i.e. prescription-disease and prescription-syndrome, was up to 90% in the linear SVM model. The macro₋averaging and micro₋averaging of the two models were around 0.92, 0.46, respectively. After AHP mapping, most of the incompatible combinations showed significant difference with other drug combinations during the clustering process in the vertical icicle, indicating that the proper machine learning algorithm can be used to lay the foundation for further exploring the combination rules in TCM and establishing more detailed drug-disease and syndrome predicting models, and provide theoretical guidance for the study of the combined use of drugs to a certain degree.