Advanced graph and sequence neural networks for molecular property prediction and drug discovery.

MOTIVATION: Properties of molecules are indicative of their functions and thus are useful in many applications. With the advances of deep-learning methods, computational approaches for predicting molecular properties are gaining increasing momentum. However, there lacks customized and advanced methods and comprehensive tools for this task currently. RESULTS: Here, we develop a suite of comprehensive machine-learning methods and tools spanning different computational models, molecular representations and loss functions for molecular property prediction and drug discovery. Specifically, we represent molecules as both graphs and sequences. Built on these representations, we develop novel deep models for learning from molecular graphs and sequences. In order to learn effectively from highly imbalanced datasets, we develop advanced loss functions that optimize areas under precision-recall curves (PRCs) and receiver operating characteristic (ROC) curves. Altogether, our work not only serves as a comprehensive tool, but also contributes toward developing novel and advanced graph and sequence-learning methodologies. Results on both online and offline antibiotics discovery and molecular property prediction tasks show that our methods achieve consistent improvements over prior methods. In particular, our methods achieve #1 ranking in terms of both ROC-AUC (area under curve) and PRC-AUC on the AI Cures open challenge for drug discovery related to COVID-19. AVAILABILITY AND IMPLEMENTATION: Our source code is released as part of the MoleculeX library (https://github.com/divelab/MoleculeX) under AdvProp. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Preparation of a DNA-Tb-MOF conjugate as a time-resolved probe for the detection of SO2 derivatives through an off-on effect.

Herein we synthesize a DNA-sensitized Tb-MOF conjugate (DNA-Tb-MOF) as a time-resolved luminescent probe to sensitively and selectively assay SO2 and their derivatives (i.e., HSO3-) through a photoluminescence off-on effect. The charge and energy transfer mechanism enables the demonstration of the effect of the photoluminescence turn-on which results from the reaction between the amino group of the DNA-Tb-MOF conjugate and SO2/HSO3-. The results demonstrate that the DNA-Tb-MOF conjugate probe can sense SO2 and their derivatives (i.e., HSO3-) with a detection limit of 0.02 ppm. Moreover, the photoluminescence off-on effect can be observed even by the naked eye.

TPP-related mitochondrial targeting copper (II) complex induces p53-dependent apoptosis in hepatoma cells through ROS-mediated activation of Drp1.

BACKGROUND: In recent years, copper complexes have gradually become the focus of potential anticancer drugs due to their available redox properties and low toxicity. In this study, a novel mitochondrion-targeting copper (II) complex, [Cu (ttpy-tpp)Br2] Br (simplified as CTB), is first synthesized by our group. CTB with tri-phenyl-phosphine (TPP), a targeting and lipophilic group, can cross the cytoplasmic and mitochondrial membranes of tumor cells. The present study aims to investigate how CTB affects mitochondrial functions and exerts its anti-tumor activity in hepatoma cells. METHODS: Multiple molecular experiments including Flow cytometry, Western blot, Immunofluorescence, Tracker staining, Transmission Electron Microscopy and Molecular docking simulation were used to elucidate the underlying mechanisms. Human hepatoma cells were subcutaneously injected into right armpit of male nude mice for evaluating the effects of CTB in vivo. RESULTS: CTB induced apoptosis via collapse of mitochondrial membrane potential (MMP), ROS production, Bax mitochondrial aggregation as well as cytochrome c release, indicating that CTB-induced apoptosis was associated with mitochondrial pathway in human hepatoma cells. Mechanistic study revealed that ROS-related mitochondrial translocation of p53 was involved in CTB-mediated apoptosis. Simultaneously, elevated mitochondrial Drp1 levels were also observed, and interruption of Drp1 activation played critical role in p53-dependent apoptosis. CTB also strongly suppressed the growth of liver cancer xenografts in vivo. CONCLUSION: In human hepatoma cells, CTB primarily induces mitochondrial dysfunction and promotes accumulation of ROS, leading to activation of Drp1. These stimulation signals accelerate mitochondrial accumulation of p53 and lead to the eventual apoptosis. Our research shows that CTB merits further evaluation as a chemotherapeutic agent for the treatment of Hepatocellular carcinoma (HCC).

Fluorometric visualization of mucin 1 glycans on cell surfaces based on rolling-mediated cascade amplification and CdTe quantum dots.

A rolling-mediated cascade (RMC) amplification strategy is described for improved visualization of profiling glycans of mucin 1 (MUC 1) on cell surfaces. CdTe quantum dots (QDs) are used as fluorescent labels. The RMC based amplification allows even distinct glycoforms of MUC1 to be visualized on the surface of MCF-7 cell via an amplified Förster resonance energy transfer (FRET) imaging strategy that works at excitation/emission wavelengths of 345/610 nm. This is achieved by utilizing antibody against MUC1 modified with the fluorescent label 7-amino-4-methylcoumarin-3-acetic acid (AMCA) as the energy donor in FRET. The QDs (used to label surface glycans) act as acceptors. N-Azidoacetylgalactosamine-Acetylated (Ac4GalNAz) as a non-natural azido sugar, can be incorporated into the glycans of the cell surface, which can promote further labeling. The method has the advantage of only requiring a small amount of non-natural sugar to be introduced in metabolic glycan labeling since too much of an artificial sugar will interfere with the physiological functions of cells. Graphical abstract Schematic for the DNA rolling-mediated cascade (RMC)-assisted metabolic labeling of cell surface glycans by using CdTe quantum dots as labels and an intramolecular amplified FRET strategy for imaging glycans on a specific glycosylated protein, MUC1.

Isorhamnetin alleviates cisplatin-induced acute kidney injury via enhancing fatty acid oxidation.

Cisplatin is an effective chemotherapy drug widely used in the treatment of various solid tumors. However, the clinical usage of cisplatin is limited by its nephrotoxicity. Isorhamnetin, a natural flavanol compound, displays remarkable pharmacological effects, including anti-inflammatory and anti-oxidation. In this study, we aimed to investigate the potential of isorhamnetin in alleviating acute kidney injury induced by cisplatin. In vitro study showed that isorhamnetin significantly suppressed the cytotoxic effects of cisplatin on human tubular epithelial cells. Furthermore, isorhamnetin exerted significantly inhibitory effects on cisplatin-induced apoptosis and inflammatory response. In acute kidney injury mice induced by a single intraperitoneal injection with 20 mg/kg cisplatin, oral administration of isorhamnetin two days before or 2 h after cisplatin injection effectively ameliorated renal function and renal tubule injury. Transcriptomics RNA-seq analysis of the mice kidney tissues suggested that isorhamnetin treatment may protect against cisplatin-induced nephrotoxicity via PGC-1α mediated fatty acid oxidation. Isorhamnetin achieved significant enhancements in the lipid clearance, ATP level, as well as the expression of PGC-1α and its downstream target genes PPARα and CPT1A, which were otherwise impaired by cisplatin. In addition, the protection effects of isorhamnetin against cisplatin-induced nephrotoxicity were abolished by a PGC-1α inhibitor, SR-18292. In conclusion, our findings indicate that isorhamnetin could protect against cisplatin-induced acute kidney injury by inducing PGC-1α-dependent reprogramming of fatty acid oxidation, which highlights the clinical potential of isorhamnetin as a therapeutic approach for the management of cisplatin-induced nephrotoxicity.

Single-cell RNA sequencing reveals cell-cell communication and potential biomarker in sepsis and septic shock patients.

BACKGROUND: Sepsis is a disease characterized by infection-induced multiorgan dysfunction, which can progress to septic shock if not promptly treated. Early identification of sepsis is crucial for its treatment. However, there are currently limited specific biomarkers for sepsis or septic shock. This study aims to identify potential biomarkers for sepsis and septic shock. METHODS: We analyzed single-cell transcriptomic data of peripheral blood mononuclear cells (PBMCs) from healthy individuals, sepsis and septic shock patients, identified differences in gene expression and cell-cell communication between different cell types during disease progression. Moreover, our analyses were further validated with flow cytometry and bulk RNA-seq data. RESULTS: Our study elucidates the alterations in cellular proportions and cell-cell communication among healthy controls, sepsis, and septic shock patients. We identified a specific augmentation in the Resistin signaling within sepsis monocytes, mediated via RETN-CAP1 ligand-receptor pairs. Additionally, we observed enhanced IL16 signaling within monocytes from septic shock patients, mediated through IL16-CD4 ligand-receptor pairs. Subsequently, we confirmed our findings by validating the increase in CAP-1+ monocytes in sepsis and IL16+ monocytes in septic shock in mouse models. And a significant upregulation of CAP-1 and IL16 was also observed in the bulk RNA-seq data from patients with sepsis and septic shock. Furthermore, we identified four distinct clusters of CD14+ monocytes, highlighting the heterogeneity of monocytes in the progress of sepsis. CONCLUSIONS: In summary, our work demonstrates changes in cell-cell communication of healthy controls, sepsis and septic shock, confirming that the molecules CAP-1 and IL16 on monocytes may serve as potential diagnostic markers for sepsis and septic shock, respectively. These findings provide new insights for early diagnosis and stratified treatment of the disease.

Unraveling the treatment effects of huanglian jiedu decoction on drug-induced liver injury based on network pharmacology, molecular docking and experimental validation.

Huanglian Jiedu Decoction (HJD) is a well-known Traditional Chinese Medicine formula that has been used for liver protection in thousands of years. However, the therapeutic effects and mechanisms of HJD in treating drug-induced liver injury (DILI) remain unknown. In this study, a total of 26 genes related to both HJD and DILI were identified, which are corresponding to a total of 41 potential active compounds in HJD. KEGG analysis revealed that Tryptophan metabolism pathway is particularly important. The overlapped genes from KEGG and GO analysis indicated the significance of CYP1A1, CYP1A2, and CYP1B1. Experimental results confirmed that HJD has a protective effect on DILI through Tryptophan metabolism pathway. In addition, the active ingredients Corymbosin, and Moslosooflavone were found to have relative strong intensity in UPLC-Q-TOF-MS/MS analysis, showing interactions with CYP1A1, CYP1A2, and CYP1B1 through molecule docking. These findings could provide insights into the treatment effects of HJD on DILI.

Unsupervised deep representation learning enables phenotype discovery for genetic association studies of brain imaging.

Understanding the genetic architecture of brain structure is challenging, partly due to difficulties in designing robust, non-biased descriptors of brain morphology. Until recently, brain measures for genome-wide association studies (GWAS) consisted of traditionally expert-defined or software-derived image-derived phenotypes (IDPs) that are often based on theoretical preconceptions or computed from limited amounts of data. Here, we present an approach to derive brain imaging phenotypes using unsupervised deep representation learning. We train a 3-D convolutional autoencoder model with reconstruction loss on 6130 UK Biobank (UKBB) participants' T1 or T2-FLAIR (T2) brain MRIs to create a 128-dimensional representation known as Unsupervised Deep learning derived Imaging Phenotypes (UDIPs). GWAS of these UDIPs in held-out UKBB subjects (n = 22,880 discovery and n = 12,359/11,265 replication cohorts for T1/T2) identified 9457 significant SNPs organized into 97 independent genetic loci of which 60 loci were replicated. Twenty-six loci were not reported in earlier T1 and T2 IDP-based UK Biobank GWAS. We developed a perturbation-based decoder interpretation approach to show that these loci are associated with UDIPs mapped to multiple relevant brain regions. Our results established unsupervised deep learning can derive robust, unbiased, heritable, and interpretable brain imaging phenotypes.

Group Contrastive Self-Supervised Learning on Graphs.

We study self-supervised learning on graphs using contrastive methods. A general scheme of prior methods is to optimize two-view representations of input graphs. In many studies, a single graph-level representation is computed as one of the contrastive objectives, capturing limited characteristics of graphs. We argue that contrasting graphs in multiple subspaces enables graph encoders to capture more abundant characteristics. To this end, we propose a group contrastive learning framework in this work. Our framework embeds the given graph into multiple subspaces, of which each representation is prompted to encode specific characteristics of graphs. To learn diverse and informative representations, we develop principled objectives that enable us to capture the relations among both intra-space and inter-space representations in groups. Under the proposed framework, we further develop an attention-based group generator to compute representations that capture different substructures of a given graph. Built upon our framework, we extend two current methods into GroupCL and GroupIG, equipped with the proposed objective. Comprehensive experimental results show our framework achieves a promising boost in performance on a variety of datasets. In addition, our qualitative results show that features generated from our representor successfully capture various specific characteristics of graphs.

Self-Supervised Learning of Graph Neural Networks: A Unified Review.

Deep models trained in supervised mode have achieved remarkable success on a variety of tasks. When labeled samples are limited, self-supervised learning (SSL) is emerging as a new paradigm for making use of large amounts of unlabeled samples. SSL has achieved promising performance on natural language and image learning tasks. Recently, there is a trend to extend such success to graph data using graph neural networks (GNNs). In this survey, we provide a unified review of different ways of training GNNs using SSL. Specifically, we categorize SSL methods into contrastive and predictive models. In either category, we provide a unified framework for methods as well as how these methods differ in each component under the framework. Our unified treatment of SSL methods for GNNs sheds light on the similarities and differences of various methods, setting the stage for developing new methods and algorithms. We also summarize different SSL settings and the corresponding datasets used in each setting. To facilitate methodological development and empirical comparison, we develop a standardized testbed for SSL in GNNs, including implementations of common baseline methods, datasets, and evaluation metrics.

Opportunities and challenges of incretin-based hypoglycemic agents treating type 2 diabetes mellitus from the perspective of physiological disposition.

The treatment of patients with diabetes mellitus, which is characterized by defective insulin secretion and/or the inability of tissues to respond to insulin, has been studied for decades. Many studies have focused on the use of incretin-based hypoglycemic agents in treating type 2 diabetes mellitus (T2DM). These drugs are classified as GLP-1 receptor agonists, which mimic the function of GLP-1, and DPP-4 inhibitors, which avoid GLP-1 degradation. Many incretin-based hypoglycemic agents have been approved and are widely used, and their physiological disposition and structural characteristics are crucial in the discovery of more effective drugs and provide guidance for clinical treatment of T2DM. Here, we summarize the functional mechanisms and other information of the drugs that are currently approved or under research for T2DM treatment. In addition, their physiological disposition, including metabolism, excretion, and potential drug-drug interactions, is thoroughly reviewed. We also discuss similarities and differences in metabolism and excretion between GLP-1 receptor agonists and DPP-4 inhibitors. This review may facilitate clinical decision making based on patients' physical conditions and the avoidance of drug-drug interactions. Moreover, the identification and development of novel drugs with appropriate physiological dispositions might be inspired.

Genetic InfoMax: Exploring Mutual Information Maximization in High-Dimensional Imaging Genetics Studies.

Genome-wide association studies (GWAS) are used to identify relationships between genetic variations and specific traits. When applied to high-dimensional medical imaging data, a key step is to extract lower-dimensional, yet informative representations of the data as traits. Representation learning for imaging genetics is largely under-explored due to the unique challenges posed by GWAS in comparison to typical visual representation learning. In this study, we tackle this problem from the mutual information (MI) perspective by identifying key limitations of existing methods. We introduce a trans-modal learning framework Genetic InfoMax (GIM), including a regularized MI estimator and a novel genetics-informed transformer to address the specific challenges of GWAS. We evaluate GIM on human brain 3D MRI data and establish standardized evaluation protocols to compare it to existing approaches. Our results demonstrate the effectiveness of GIM and a significantly improved performance on GWAS.

Augmented Equivariant Attention Networks for Microscopy Image Transformation.

It is time-consuming and expensive to take high-quality or high-resolution electron microscopy (EM) and fluorescence microscopy (FM) images. Taking these images could be even invasive to samples and may damage certain subtleties in the samples after long or intense exposures, often necessary for achieving high-quality or high-resolution in the first place. Advances in deep learning enable us to perform various types of microscopy image-to-image transformation tasks such as image denoising, super-resolution, and segmentation that computationally produce high-quality images from the physically acquired low-quality ones. When training image-to-image transformation models on pairs of experimentally acquired microscopy images, prior models suffer from performance loss due to their inability to capture inter-image dependencies and common features shared among images. Existing methods that take advantage of shared features in image classification tasks cannot be properly applied to image transformation tasks because they fail to preserve the equivariance property under spatial permutations, something essential in image-to-image transformation. To address these limitations, we propose the augmented equivariant attention networks (AEANets) with better capability to capture inter-image dependencies, while preserving the equivariance property. The proposed AEANets captures inter-image dependencies and shared features via two augmentations on the attention mechanism, which are the shared references and the batch-aware attention during training. We theoretically derive the equivariance property of the proposed augmented attention model and experimentally demonstrate its consistent superiority in both quantitative and visual results over the baseline methods.

Air pollution in China: Status and spatiotemporal variations.

In recent years, China has experienced severe and persistent air pollution associated with rapid urbanization and climate change. Three years' time series (January 2014 to December 2016) concentrations data of air pollutants including particulate matter (PM2.5 and PM10) and gaseous pollutants (SO2, NO2, CO, and O3) from over 1300 national air quality monitoring sites were studied to understand the severity of China's air pollution. In 2014 (2015, 2016), annual population-weighted-average (PWA) values in China were 65.8 (55.0, 50.7) µg m-3 for PM2.5, 107.8 (91.1, 85.7) µg m-3 for PM10, 54.8 (56.2, 57.2) µg m-3 for O3_8 h, 39.6 (33.3, 33.4) µg m-3 for NO2, 34.1 (26, 21.9) µg m-3 for SO2, 1.2 (1.1, 1.1) mg m-3 for CO, and 0.60 (0.59, 0.58) for PM2.5/PM10, respectively. In 2014 (2015, 2016), 7% (14%, 19%), 17% (27%, 34%), 51% (67%, 70%) and 88% (97%, 98%) of the population in China lived in areas that meet the level of annual PM2.5, PM10, NO2, and SO2 standard metrics from Chinese Ambient Air Quality Standards-Grade II. The annual PWA concentrations of PM2.5, PM10, O3_8 h, NO2, SO2, CO in the Northern China are about 40.4%, 58.9%, 5.9%, 24.6%, 96.7%, and 38.1% higher than those in Southern China, respectively. Though the air quality has been improving recent years, PM2.5 pollution in wintertime is worsening, especially in the Northern China. The complex air pollution caused by PM and O3 (the third frequent major pollutant) is an emerging problem that threatens the public health, especially in Chinese mega-city clusters. NOx controls were more beneficial than SO2 controls for improvement of annual PM air quality in the northern China, central, and southwest regions. Future epidemiologic studies are urgently required to estimate the health impacts associated with multi-pollutants exposure, and revise more scientific air quality index standards.