A flexible data-free framework for structure-based de novo drug design with reinforcement learning.

Contemporary structure-based molecular generative methods have demonstrated their potential to model the geometric and energetic complementarity between ligands and receptors, thereby facilitating the design of molecules with favorable binding affinity and target specificity. Despite the introduction of deep generative models for molecular generation, the atom-wise generation paradigm that partially contradicts chemical intuition limits the validity and synthetic accessibility of the generated molecules. Additionally, the dependence of deep learning models on large-scale structural data has hindered their adaptability across different targets. To overcome these challenges, we present a novel search-based framework, 3D-MCTS, for structure-based de novo drug design. Distinct from prevailing atom-centric methods, 3D-MCTS employs a fragment-based molecular editing strategy. The fragments decomposed from small-molecule drugs are recombined under predefined retrosynthetic rules, offering improved drug-likeness and synthesizability, overcoming the inherent limitations of atom-based approaches. Leveraging multi-threaded parallel simulations combined with a real-time energy constraint-based pruning strategy, 3D-MCTS achieves remarkable efficiency. At a fixed computational cost, it outperforms other state-of-the-art (SOTA) methods by producing molecules with enhanced binding affinity. Furthermore, its fragment-based approach ensures the generation of more dependable binding conformations, exhibiting a success rate 43.6% higher than that of other SOTAs. This advantage becomes even more pronounced when handling targets that significantly deviate from the training dataset. 3D-MCTS is capable of achieving thirty times more hits with high binding affinity than traditional virtual screening methods, which demonstrates the superior ability of 3D-MCTS to explore chemical space. Moreover, the flexibility of our framework makes it easy to incorporate domain knowledge during the process, thereby enabling the generation of molecules with desirable pharmacophores and enhanced binding affinity. The adaptability of 3D-MCTS is further showcased in metalloprotein applications, highlighting its potential across various drug design scenarios.

Multi-Dimensional Comparison of the Impact Mechanism of the Self-Rated Health Status of Urban and Rural Residents in Chinese Social Environments.

Self-rated health status (SRHS) reflects individuals' social environment, and the difference between urban and rural areas in China further highlights the impact of social environment on health. This paper aimed to systematically analyze and compare the impact mechanism of the SRHS of urban and rural residents from multiple dimensions, i.e., time, space, and scale. Drawing on data from the Chinese General Social Survey (CGSS) and China Statistical Yearbook, we used spatial, cross, and HLM analyses. Results indicate that: (1) From 2010 to 2017, the overall SRHS level of Chinese residents gradually declined; the gradient pattern of east, middle, and west became more marked, and the health level in rural areas generally fell behind that of urban areas. (2) The focus of SRHS moved toward mental health, and people's perceptions of the social environment gradually became a key factor affecting health. (3) In the long term, the gradient allocation of medical service resources could narrow the gap between urban and rural areas to comprehensively improve regional health levels.

Proteome-Wide Profiling of the Covalent-Druggable Cysteines with a Structure-Based Deep Graph Learning Network.

Covalent ligands have attracted increasing attention due to their unique advantages, such as long residence time, high selectivity, and strong binding affinity. They also show promise for targets where previous efforts to identify noncovalent small molecule inhibitors have failed. However, our limited knowledge of covalent binding sites has hindered the discovery of novel ligands. Therefore, developing in silico methods to identify covalent binding sites is highly desirable. Here, we propose DeepCoSI, the first structure-based deep graph learning model to identify ligandable covalent sites in the protein. By integrating the characterization of the binding pocket and the interactions between each cysteine and the surrounding environment, DeepCoSI achieves state-of-the-art predictive performances. The validation on two external test sets which mimic the real application scenarios shows that DeepCoSI has strong ability to distinguish ligandable sites from the others. Finally, we profiled the entire set of protein structures in the RCSB Protein Data Bank (PDB) with DeepCoSI to evaluate the ligandability of each cysteine for covalent ligand design, and made the predicted data publicly available on website.

Double-Balanced Loss for Imbalanced Colorectal Lesion Classification.

Colorectal cancer has a high incidence rate in all countries around the world, and the survival rate of patients is improved by early detection. With the development of object detection technology based on deep learning, computer-aided diagnosis of colonoscopy medical images becomes a reality, which can effectively reduce the occurrence of missed diagnosis and misdiagnosis. In medical image recognition, the assumption that training samples follow independent identical distribution (IID) is the key to the high accuracy of deep learning. However, the classification of medical images is unbalanced in most cases. This paper proposes a new loss function named the double-balanced loss function for the deep learning model, to improve the impact of datasets on classification accuracy. It introduces the effects of sample size and sample difficulty to the loss calculation and deals with both sample size imbalance and sample difficulty imbalance. And it combines with deep learning to build the medical diagnosis model for colorectal cancer. Experimentally verified by three colorectal white-light endoscopic image datasets, the double-balanced loss function proposed in this paper has better performance on the imbalance classification problem of colorectal medical images.

Boosting Protein-Ligand Binding Pose Prediction and Virtual Screening Based on Residue-Atom Distance Likelihood Potential and Graph Transformer.

The past few years have witnessed enormous progress toward applying machine learning approaches to the development of protein-ligand scoring functions. However, the robust performance and wide applicability of scoring functions remain a big challenge for increasing the success rate of docking-based virtual screening. Herein, a novel scoring function named RTMScore was developed by introducing a tailored residue-based graph representation strategy and several graph transformer layers for the learning of protein and ligand representations, followed by a mixture density network to obtain residue-atom distance likelihood potential. Our approach was resolutely validated on the CASF-2016 benchmark, and the results indicate that RTMScore can outperform almost all of the other state-of-the-art methods in terms of both the docking and screening powers. Further evaluation confirms the robustness of our approach that can not only retain its docking power on cross-docked poses but also achieve improved performance as a rescoring tool in larger-scale virtual screening.

Organic Compound Synthetic Accessibility Prediction Based on the Graph Attention Mechanism.

Accurate estimation of the synthetic accessibility of small molecules is needed in many phases of drug discovery. Several expert-crafted scoring methods and descriptor-based quantitative structure-activity relationship (QSAR) models have been developed for synthetic accessibility assessment, but their practical applications in drug discovery are still quite limited because of relatively low prediction accuracy and poor model interpretability. In this study, we proposed a data-driven interpretable prediction framework called GASA (Graph Attention-based assessment of Synthetic Accessibility) to evaluate the synthetic accessibility of small molecules by distinguishing compounds to be easy- (ES) or hard-to-synthesize (HS). GASA is a graph neural network (GNN) architecture that makes self-feature deduction by applying an attention mechanism to automatically capture the most important structural features related to synthetic accessibility. The sampling around the hypothetical classification boundary was used to improve the ability of GASA to distinguish structurally similar molecules. GASA was extensively evaluated and compared with two descriptor-based machine learning methods (random forest, RF; eXtreme gradient boosting, XGBoost) and four existing scores (SYBA: SYnthetic Bayesian Accessibility; SCScore: Synthetic Complexity score; RAscore: Retrosynthetic Accessibility score; SAscore: Synthetic Accessibility score). Our analysis demonstrates that GASA achieved remarkable performance in distinguishing similar molecules compared with other methods and had a broader applicability domain. In addition, we show how GASA learns the important features that affect molecular synthetic accessibility by assigning attention weights to different atoms. An online prediction service for GASA was offered at http://cadd.zju.edu.cn/gasa/.

Structural view on the role of the TRD loop in regulating DNMT3A activity: a molecular dynamics study.

DNA methyltransferase 3A (DNMT3A) has been regarded as a potential epigenetic target for the development of cancer therapeutics. A number of DNMT3A inhibitors have been reported, but most of them do not have good potency, high selectivity and/or low cytotoxicity. It has been suggested that a non-conserved region around the target recognition domain (TRD) loop is implicated in the DNMT3A activity under the allosteric regulation of the ATRX-DNMT3-DNMT3L (ADD) domain, but the molecular mechanism of the regulation of the TRD loop on the DNMT3A activity needs to be elucidated. In this study, based on the reported crystal structures, the dynamics of the TRD loop in different multimerization with/without the bound guest molecule, namely the ADD domain or the DNA molecule, was investigated using conventional molecular dynamics (MD) and umbrella sampling simulations. The simulation results illustrate that the TRD loop exhibits relatively higher flexibility than the other components in the whole catalytic domain (CD), which could be well stabilized into different local minima through the binding with either the ADD domain or the DNA molecule by forming tight hydrogen-bond and salt-bridge networks involving distinct residues. Moreover, the movement of the TRD loop away from the catalytic loop upon activation could be triggered simply by the detachment of the ADD domain, but not necessarily induced by the ADD domain relocation on the CD. All these dynamic structural details could be a supplement to the previously reported crystal structure, which underlines the importance of the structural flexibility for the critical residues in the TRD loop, arousing more interest in the rational design of novel DNMT3A inhibitors targeting this region.

White-Light Endoscopic Colorectal Lesion Detection Based on Improved YOLOv5.

As an effective tool for colorectal lesion detection, it is still difficult to avoid the phenomenon of missed and false detection when using white-light endoscopy. In order to improve the lesion detection rate of colorectal cancer patients, this paper proposes a real-time lesion diagnosis model (YOLOv5x-CG) based on YOLOv5 improvement. In this diagnostic model, colorectal lesions were subdivided into three categories: micropolyps, adenomas, and cancer. In the course of convolutional network training, Mosaic data enhancement strategy was used to improve the detection rate of small target polyps. At the same time, coordinate attention (CA) mechanism was introduced to take into account channel and location information in the network, so as to realize the effective extraction of three kinds of pathological features. The Ghost module was also used to generate more feature maps through linear processing, which reduces the stress of learning model parameters and speeds up detection. The experimental results show that the lesion diagnosis model proposed in this paper has a more rapid and accurate lesion detection ability, and the AP value of polyps, adenomas, and cancer is 0.923, 0.955, and 0.87, and mAP@50 is 0.916.

Spiroarborin, an ent-Clerodane Homodimer from Callicarpa arborea as an Inhibitor of the Eleven-Nineteen Leukemia (ENL) Protein by Targeting the YEATS Domain.

A spiro ent-clerodane homodimer with a rare 6/6/6/6/6-fused pentacyclic scaffold, spiroarborin (1), together with four new monomeric analogues (2-5), were isolated from Callicarpa arborea. Their structures were elucidated by comprehensive spectroscopic data analysis, quantum-chemical calculations, and X-ray diffraction. A plausible biosynthetic pathway of 1 was proposed, and a biomimetic synthesis of its derivative was accomplished. Compound 1 showed a potent inhibitory effect by directly binding to the YEATS domain of the 11-19 leukemia (ENL) protein with an IC50 value of 7.3 µM. This gave a KD value of 5.0 µM, as recorded by a surface plasmon resonance binding assay.

The impact of cross-docked poses on performance of machine learning classifier for protein-ligand binding pose prediction.

Structure-based drug design depends on the detailed knowledge of the three-dimensional (3D) structures of protein-ligand binding complexes, but accurate prediction of ligand-binding poses is still a major challenge for molecular docking due to deficiency of scoring functions (SFs) and ignorance of protein flexibility upon ligand binding. In this study, based on a cross-docking dataset dedicatedly constructed from the PDBbind database, we developed several XGBoost-trained classifiers to discriminate the near-native binding poses from decoys, and systematically assessed their performance with/without the involvement of the cross-docked poses in the training/test sets. The calculation results illustrate that using Extended Connectivity Interaction Features (ECIF), Vina energy terms and docking pose ranks as the features can achieve the best performance, according to the validation through the random splitting or refined-core splitting and the testing on the re-docked or cross-docked poses. Besides, it is found that, despite the significant decrease of the performance for the threefold clustered cross-validation, the inclusion of the Vina energy terms can effectively ensure the lower limit of the performance of the models and thus improve their generalization capability. Furthermore, our calculation results also highlight the importance of the incorporation of the cross-docked poses into the training of the SFs with wide application domain and high robustness for binding pose prediction. The source code and the newly-developed cross-docking datasets can be freely available at https://github.com/sc8668/ml_pose_prediction and https://zenodo.org/record/5525936 , respectively, under an open-source license. We believe that our study may provide valuable guidance for the development and assessment of new machine learning-based SFs (MLSFs) for the predictions of protein-ligand binding poses.

Origin and Enrichment of Vanadium in the Lower Cambrian Black Shales, South China.

The Lower Cambrian black shales of the Sansui vanadium deposits, located in South China, host a thin accumulation of Ni, Mo, V, and platinum group of elements (PGE). However, among them, the origin of V-bearing deposits remains controversial. To characterize the enrichment process of V-bearing deposits, samples of the mineralized layer and surrounding rocks from the Sansui area, South China, were investigated through bulk geochemical analysis and scanning electron microscopy (SEM) and energy dispersive spectrometry (EDS) analyses. There is a consistency in the change curves of Mo, Ni, and V from the Sansui V deposits, but the contents of elements show a great difference. This means the strong similarities in the metal sources but a difference in enrichment factors of Mo, Ni, and V. The presence of the tuff and the barite layer in the Sansui V deposits indicates that the formation of the associated V deposits was closely related to either a volcanic or hydrothermal activity. Analysis of geochemistry and sedimentation suggests a hydrothermal source of the metals, where the mineralization of V is related to clay and organic matter. Phosphorus nodules were observed at all sites of black shale V deposits in early Cambrian and were most likely related to the upwelling currents during the depositional period. The comparison with the Ni-Mo deposits highlights a stronger enrichment of clay in the V deposits. The V deposits are located in the lower part of the continental slope. Both organic matter and clay minerals are abundant in the Sansui deposits. However, the isomorphism of V-Al is promoted by the hydrothermal activity and suggests that the origin of V deposits is a multistage process.

Novel clerodane-type diterpenoid Cintelactone A suppresses lipopolysaccharide -induced inflammation by promoting ubiquitination, proteasomal degradation of TRAF6.

Cellular inflammation is the underlying cause of several diseases and development of a safe and effective anti-inflammatory drug is need-of-the hour for treatment of diseases like lung inflammation. Callicarpa integerrima Champ. is a well-known herbal medicine with hemostatic and anti-inflammatory functions. However, the exact ingredient exhibiting anti-inflammatory activity in C. integerrima Champ. is largely unknown. Here, we first isolated, purified and characterized a novel clerodane-type diterpenoid Cintelactone A (CA) from C. integerrima Champ. We demonstrated that CA could significantly inhibit lipopolysaccharide (LPS)-induced pro-inflammatory cytokines and mediators production both in mouse peritoneal macrophages and THP1 cells. Consistently, CA also relieved inflammation and reduced LPS-induced lung injury in mice. We systematically elucidated the mechanism of action as well. CA interacted with Arg78 of tumor necrosis factor receptor-associated factor 6 (TRAF6) by hydrogen bonding. It further promoted the K48-linked ubiquitination and proteasomal degradation of TRAF6, and suppressed the activation of NF-κB and MAPKs signaling pathways. Collectively, our study reveals that new clerodane-type diterpenoid CA suppresses LPS-induced inflammation by promoting TRAF6 degradation, suggesting that CA as the potential therapeutic candidate for the treatment of inflammation associated diseases.

CovalentInDB: a comprehensive database facilitating the discovery of covalent inhibitors.

Inhibitors that form covalent bonds with their targets have traditionally been considered highly adventurous due to their potential off-target effects and toxicity concerns. However, with the clinical validation and approval of many covalent inhibitors during the past decade, design and discovery of novel covalent inhibitors have attracted increasing attention. A large amount of scattered experimental data for covalent inhibitors have been reported, but a resource by integrating the experimental information for covalent inhibitor discovery is still lacking. In this study, we presented Covalent Inhibitor Database (CovalentInDB), the largest online database that provides the structural information and experimental data for covalent inhibitors. CovalentInDB contains 4511 covalent inhibitors (including 68 approved drugs) with 57 different reactive warheads for 280 protein targets. The crystal structures of some of the proteins bound with a covalent inhibitor are provided to visualize the protein-ligand interactions around the binding site. Each covalent inhibitor is annotated with the structure, warhead, experimental bioactivity, physicochemical properties, etc. Moreover, CovalentInDB provides the covalent reaction mechanism and the corresponding experimental verification methods for each inhibitor towards its target. High-quality datasets are downloadable for users to evaluate and develop computational methods for covalent drug design. CovalentInDB is freely accessible at http://cadd.zju.edu.cn/cidb/.

PROTAC-DB: an online database of PROTACs.

Proteolysis-targeting chimeras (PROTACs), which selectively degrade targeted proteins by the ubiquitin-proteasome system, have emerged as a novel therapeutic technology with potential advantages over traditional inhibition strategies. In the past few years, this technology has achieved substantial progress and two PROTACs have been advanced into phase I clinical trials. However, this technology is still maturing and the design of PROTACs remains a great challenge. In order to promote the rational design of PROTACs, we present PROTAC-DB, a web-based open-access database that integrates structural information and experimental data of PROTACs. Currently, PROTAC-DB consists of 1662 PROTACs, 202 warheads (small molecules that target the proteins of interest), 65 E3 ligands (small molecules capable of recruiting E3 ligases) and 806 linkers, as well as their chemical structures, biological activities, and physicochemical properties. Except the biological activities of warheads and E3 ligands, PROTAC-DB also provides the degradation capacities, binding affinities and cellular activities for PROTACs. PROTAC-DB can be queried with two general searching approaches: text-based (target name, compound name or ID) and structure-based. In addition, for the convenience of users, a filtering tool for the searching results based on the physicochemical properties of compounds is also offered. PROTAC-DB is freely accessible at http://cadd.zju.edu.cn/protacdb/.

Application of Deep Learning for Early Screening of Colorectal Precancerous Lesions under White Light Endoscopy.

METHODS: We collected and sorted out the white light endoscopic images of some patients undergoing colonoscopy. The convolutional neural network model is used to detect whether the image contains lesions: CRC, colorectal adenoma (CRA), and colorectal polyps. The accuracy, sensitivity, and specificity rates are used as indicators to evaluate the model. Then, the instance segmentation model is used to locate and classify the lesions on the images containing lesions, and mAP (mean average precision), AP50, and AP75 are used to evaluate the performance of an instance segmentation model. RESULTS: In the process of detecting whether the image contains lesions, we compared ResNet50 with the other four models, that is, AlexNet, VGG19, ResNet18, and GoogLeNet. The result is that ResNet50 performs better than several other models. It scored an accuracy of 93.0%, a sensitivity of 94.3%, and a specificity of 90.6%. In the process of localization and classification of the lesion in images containing lesions by Mask R-CNN, its mAP, AP50, and AP75 were 0.676, 0.903, and 0.833, respectively. CONCLUSION: We developed and compared five models for the detection of lesions in white light endoscopic images. ResNet50 showed the optimal performance, and Mask R-CNN model could be used to locate and classify lesions in images containing lesions.

Callicarpins, Two Classes of Rearranged ent-Clerodane Diterpenoids from Callicarpa Plants Blocking NLRP3 Inflammasome-Induced Pyroptosis.

Callicarpins A-D (1-4), possessing an unprecedented A-homoent-clerodane scaffold with a bicyclo[5.4.0]undecane ring system, and callicarpins E-G (5-7), with 5/6-fused ent-clerodane diterpenoid skeletons, were isolated from Callicarpaarborea and C. integerrim. Their structures were elucidated by comprehensive spectroscopic data, X-ray crystal diffraction, chemical derivatization, and electronic circular dichroism (ECD) data. Putative biosynthetic pathways for these callicarpins are proposed. Compounds 2, 3b, and 6-8 showed potent inhibitory effects against the NLRP3 inflammasome with IC50 values from 1.4 to 5.3 µM, and 2 significantly blocked NLRP3 inflammasome-induced pyroptosis by inhibiting Casp-1 activation and IL-1ß secretion in J774A.1 cells.

Luteolin-7-methylether from Leonurus japonicus inhibits estrogen biosynthesis in human ovarian granulosa cells by suppression of aromatase (CYP19).

Leonurus japonicus (motherwort) has been widely used to treat gynecological disorders, in which estrogen is often dysregulated, for a long time in China and other Asian countries. However, the chemical constituents and mechanisms underlying the activity of this medicinal plant are not fully understood. Seventeen of forty-six tested natural products from L. japonicus showed stimulatory or inhibitory effects on estrogen biosynthesis with different potency in human ovarian granulosa-like KGN cells. Luteolin-7-methylether (XLY29) potently inhibited 17ß-estradiol production (IC50: 5.213 µM) by decreasing the expression of aromatase, the only enzyme in vertebrates that catalyzes the biosynthesis of estrogens, but had no effect on the catalytic activity of aromatase. XLY29 decreased the expression of aromatase promoter I.3/II, and suppressed the phosphorylation of cAMP response element-binding protein. XLY29 potently inhibited phosphorylation of p38 mitogen-activated protein kinase and AKT but had no effect on phosphorylation of extracellular signal-regulated kinase and c-Jun N-terminal kinase. XLY29 also decreased the serum 17ß-estradiol level and disturbed estrous cycle in mice. These results suggest that modulation of estrogen biosynthesis is a novel effect of L. japonicus, and XLY29 warrants further investigation as a new therapeutic means for the treatment of estrogen-related diseases.

Comprehensive Evaluation of Fourteen Docking Programs on Protein-Peptide Complexes.

A large number of protein-protein interactions (PPIs) are mediated by the interactions between proteins and peptide segments binding partners, and therefore determination of protein-peptide interactions (PpIs) is quite crucial to elucidate important biological processes and design peptides or peptidomimetic drugs that can modulate PPIs. Nowadays, as a powerful computation tool, molecular docking has been widely utilized to predict the binding structures of protein-peptide complexes. However, although a number of docking programs have been available, the systematic study on the assessment of their performance for PpIs has never been reported. In this study, a benchmark data set called PepSet consisting of 185 protein-peptide complexes with peptide length ranging from 5 to 20 residues was employed to evaluate the performance of 14 docking programs, including three protein-protein docking programs (ZDOCK, FRODOCK, and HawkDock), three small molecule docking programs (GOLD, Surflex-Dock, and AutoDock Vina), and eight protein-peptide docking programs (GalaxyPepDock, MDockPeP, HPEPDOCK, CABS-dock, pepATTRACT, DINC, AutoDock CrankPep (ADCP), and HADDOCK peptide docking). A new evaluation parameter, named IL_RMSD, was proposed to measure the docking accuracy with fnat (the fraction of native contacts). In global docking, HPEPDOCK performs the best for the entire data set and yields the success rates of 4.3%, 24.3%, and 55.7% at the top 1, 10, and 100 levels, respectively. In local docking, overall, ADCP achieves the best predictions and reaches the success rates of 11.9%, 37.3%, and 70.3% at the top 1, 10, and 100 levels, respectively. It is expected that our work can provide some helpful insights into the selection and development of improved docking programs for PpIs. The benchmark data set is freely available at http://cadd.zju.edu.cn/pepset/.

Combined strategies in structure-based virtual screening.

The identification and optimization of lead compounds are inalienable components in drug design and discovery pipelines. As a powerful computational approach for the identification of hits with novel structural scaffolds, structure-based virtual screening (SBVS) has exhibited a remarkably increasing influence in the early stages of drug discovery. During the past decade, a variety of techniques and algorithms have been proposed and tested with different purposes in the scope of SBVS. Although SBVS has been a common and proven technology, it still shows some challenges and problems that are needed to be addressed, where the negative influence regardless of protein flexibility and the inaccurate prediction of binding affinity are the two major challenges. Here, focusing on these difficulties, we summarize a series of combined strategies or workflows developed by our group and others. Furthermore, several representative successful applications from recent publications are also discussed to demonstrate the effectiveness of the combined SBVS strategies in drug discovery campaigns.

Triterpenoids from Ganoderma gibbosum: A Class of Sensitizers of FLC-Resistant Candida albicans to Fluconazole.

Fungal drug resistance is a major health threat, and reports of clinical resistance worldwide are becoming increasingly common. In a research program to discover new molecules to help overcome this problem, 14 new lanostane-type triterpenoids, gibbosicolids A-G (2-8) and gibbosic acids I-O (9-15), were isolated from the fruiting bodies of Ganoderma gibbosum, along with seven known triterpenoid derivatives. These compounds featured high levels of oxidation, epimerization, and γ-lactonization. Structures were elucidated by comprehensive spectroscopic analyses and HRMS data. Absolute configurations were assigned based on quantum chemical calculations, including calculated chemical shift with DP4+ analysis, coupling constants, and electronic circular dichroism (ECD) methods. Results show that the calculated NMR with DP4+ analysis could not reliably establish the overall spatial configuration of molecules possessing independent and free-rotational stereoclusters. All these compounds significantly increased the sensitivity of fluconazole (FLC)-resistant C. albicans to FLC. Compounds 2, 5, 9, 12, 16, 17, and 21 exhibited strong antifungal activity against FLC-resistant C. albicans when combined with FLC, with MIC50 values ranging from 3.8 to 8.8 µg/mL.