R2-DDI: relation-aware feature refinement for drug-drug interaction prediction.

Precisely predicting the drug-drug interaction (DDI) is an important application and host research topic in drug discovery, especially for avoiding the adverse effect when using drug combination treatment for patients. Nowadays, machine learning and deep learning methods have achieved great success in DDI prediction. However, we notice that most of the works ignore the importance of the relation type when building the DDI prediction models. In this work, we propose a novel R$^2$-DDI framework, which introduces a relation-aware feature refinement module for drug representation learning. The relation feature is integrated into drug representation and refined in the framework. With the refinement features, we also incorporate the consistency training method to regularize the multi-branch predictions for better generalization. Through extensive experiments and studies, we demonstrate our R$^2$-DDI approach can significantly improve the DDI prediction performance over multiple real-world datasets and settings, and our method shows better generalization ability with the help of the feature refinement design.

Breaking the barriers of data scarcity in drug-target affinity prediction.

Accurate prediction of drug-target affinity (DTA) is of vital importance in early-stage drug discovery, facilitating the identification of drugs that can effectively interact with specific targets and regulate their activities. While wet experiments remain the most reliable method, they are time-consuming and resource-intensive, resulting in limited data availability that poses challenges for deep learning approaches. Existing methods have primarily focused on developing techniques based on the available DTA data, without adequately addressing the data scarcity issue. To overcome this challenge, we present the Semi-Supervised Multi-task training (SSM) framework for DTA prediction, which incorporates three simple yet highly effective strategies: (1) A multi-task training approach that combines DTA prediction with masked language modeling using paired drug-target data. (2) A semi-supervised training method that leverages large-scale unpaired molecules and proteins to enhance drug and target representations. This approach differs from previous methods that only employed molecules or proteins in pre-training. (3) The integration of a lightweight cross-attention module to improve the interaction between drugs and targets, further enhancing prediction accuracy. Through extensive experiments on benchmark datasets such as BindingDB, DAVIS and KIBA, we demonstrate the superior performance of our framework. Additionally, we conduct case studies on specific drug-target binding activities, virtual screening experiments, drug feature visualizations and real-world applications, all of which showcase the significant potential of our work. In conclusion, our proposed SSM-DTA framework addresses the data limitation challenge in DTA prediction and yields promising results, paving the way for more efficient and accurate drug discovery processes.

The synergistic effect of residues 32T and 550L in the PA protein of H5 subtype avian influenza virus contributes to viral pathogenicity in mice.

The avian influenza virus (AIV) PA protein contributes to viral replication and pathogenicity; however, its interaction with innate immunity is not well understood. Here, we report that the H5 subtype AIV PA protein strongly suppresses host antiviral defense by interacting with and degrading a key protein in interferon (IFN) signaling, Janus kinase 1 (JAK1). Specifically, the AIV PA protein catalyzes the K48-linked polyubiquitination and degradation of JAK1 at lysine residue 249. Importantly, the AIV PA protein harboring 32T/550L degrades both avian and mammalian JAK1, while the AIV PA protein with residues 32M/550I degrades avian JAK1 only. Furthermore, the residues 32T/550L in PA protein confer optimum polymerase activity and AIV growth in mammalian cells. Notably, the replication and virulence of the AIV PA T32M/L550I mutant are attenuated in infected mice. Collectively, these data reveal an interference role for H5 subtype AIV PA protein in host innate immunity, which can be targeted for the development of specific and effective anti-influenza therapeutics.

Biosafety of human environments can be supported by effective use of renewable biomass.

Preventing pathogenic viral and bacterial transmission in the human environment is critical, especially in potential outbreaks that may be caused by the release of ancient bacteria currently trapped in the permafrost. Existing commercial disinfectants present issues such as a high carbon footprint. This study proposes a sustainable alternative, a bioliquid derived from biomass prepared by hydrothermal liquefaction. Results indicate a high inactivation rate of pathogenic virus and bacteria by the as-prepared bioliquid, such as up to 99.99% for H1N1, H5N1, H7N9 influenza A virus, and Bacillus subtilis var. niger spores and 99.49% for Bacillus anthracis Inactivation of Escherichia coli and Staphylococcus epidermidis confirmed that low-molecular-weight and low-polarity compounds in bioliquid are potential antibacterial components. High temperatures promoted the production of antibacterial substances via depolymerization and dehydration reactions. Moreover, bioliquid was innoxious as confirmed by the rabbit skin test, and the cost per kilogram of the bioliquid was $0.04427, which is notably lower than that of commercial disinfectants. This study demonstrates the potential of biomass to support our biosafety with greater environmental sustainability.

An Electrochemiluminescent Sensor Based on Glycosyl Imprinting and Aptamer for the Detection of Cancer-Related Extracellular Vesicles.

Cancer-related extracellular vesicles (EVs) are considered important biomarkers for cancer diagnosis because they can convey a large amount of information about tumor cells. In order to detect cancer-related EVs efficiently, an electrochemiluminescence (ECL) sensor for the specific identification and highly sensitive detection of EVs in the plasma of cancer patients was constructed based on dual recognitions by glycosyl-imprinted polymer (GIP) and aptamer. The characteristic glycosyl Neu5Ac-α-(2,6)-Gal-ß-(1-4)-GlcNAc trisaccharide on the surface of EVs was used as a template molecule and 3-aminophenylboronic acid as a functional monomer to form a glycosyl-imprinted polymer by electropolymerization. After glycosyl elution, the imprinted film specifically recognized and adsorbed the EVs in the sample, and then the CD63 aptamer-bipyridine ruthenium (Aptamer-Ru(bpy)) was added to combine with the CD63 glycoprotein on the extracellular vesicle's surface, thus providing secondary recognition of the EVs. Finally, the EVs were quantitatively detected according to the ECL signal produced by the labeled bipyridine ruthenium. When more EVs were captured by the imprinted film, more probes were obtained after incubation, and the ECL signal was stronger. Under the optimized conditions, the ECL signal showed a good linear relationship with the concentration of EVs in the range of 9.5 × 102 to 9.5 × 107 particles/mL, and the limit of detection was 641 particles/mL. The GIP sensor can discriminate between the EV contents of cancer patients and healthy controls with high accuracy. Because of its affordability, high sensitivity, and ease of use, it is anticipated to be employed for cancer early detection and diagnosis.

Recombinant hirudin and PAR-1 regulate macrophage polarisation status in diffuse large B-cell lymphoma.

BACKGROUND: Diffuse large B-cell lymphoma (DLBCL) is a malignant tumour. Although some standard therapies have been established to improve the cure rate, they remain ineffective for specific individuals. Therefore, it is meaningful to find more novel therapeutic approaches. Macrophage polarisation is extensively involved in the process of tumour development. Recombinant hirudin (rH) affects macrophages and has been researched frequently in clinical trials lately. Our article validated the regulatory role of rH in macrophage polarisation and the mechanism of PAR-1 by collecting clinical samples and subsequently establishing a cellular model to provide a scientifically supported perspective for discovering new therapeutic approaches. METHOD: We assessed the expression of macrophage polarisation markers, cytokines and PAR-1 in clinical samples. We established a cell model by co-culture with THP-1 and OCI-Ly10 cell. We determined the degree of cell polarisation and expression of validation cytokines by flow cytometry, ELISA, and RT-qPCR to confirm the success of the cell model. Subsequently, different doses of rH were added to discover the function of rH on cell polarisation. We confirmed the mechanism of PAR-1 in macrophage polarisation by transfecting si-PAR-1 and pcDNA3.1-PAR-1. RESULTS: We found higher expression of M2 macrophage markers (CD163 + CMAF+) and PAR-1 in 32 DLBCL samples. After inducing monocyte differentiation into M0 macrophages and co-culturing with OCI-Ly10 lymphoma cells, we found a trend of these expressions in the cell model consistent with the clinical samples. Subsequently, we discovered that rH promotes the polarisation of M1 macrophages but inhibits the polarisation of M2 macrophages. We also found that PAR-1 regulates macrophage polarisation, inhibiting cell proliferation, migration, invasion and angiogenic capacity. CONCLUSION: rH inhibits macrophage polarisation towards the M2 type and PAR-1 regulates polarisation, proliferation, migration, invasion, and angiogenesis of DLBCL-associated macrophages.

BioGPT: generative pre-trained transformer for biomedical text generation and mining.

Pre-trained language models have attracted increasing attention in the biomedical domain, inspired by their great success in the general natural language domain. Among the two main branches of pre-trained language models in the general language domain, i.e. BERT (and its variants) and GPT (and its variants), the first one has been extensively studied in the biomedical domain, such as BioBERT and PubMedBERT. While they have achieved great success on a variety of discriminative downstream biomedical tasks, the lack of generation ability constrains their application scope. In this paper, we propose BioGPT, a domain-specific generative Transformer language model pre-trained on large-scale biomedical literature. We evaluate BioGPT on six biomedical natural language processing tasks and demonstrate that our model outperforms previous models on most tasks. Especially, we get 44.98%, 38.42% and 40.76% F1 score on BC5CDR, KD-DTI and DDI end-to-end relation extraction tasks, respectively, and 78.2% accuracy on PubMedQA, creating a new record. Our case study on text generation further demonstrates the advantage of BioGPT on biomedical literature to generate fluent descriptions for biomedical terms.

SPRoBERTa: protein embedding learning with local fragment modeling.

Well understanding protein function and structure in computational biology helps in the understanding of human beings. To face the limited proteins that are annotated structurally and functionally, the scientific community embraces the self-supervised pre-training methods from large amounts of unlabeled protein sequences for protein embedding learning. However, the protein is usually represented by individual amino acids with limited vocabulary size (e.g. 20 type proteins), without considering the strong local semantics existing in protein sequences. In this work, we propose a novel pre-training modeling approach SPRoBERTa. We first present an unsupervised protein tokenizer to learn protein representations with local fragment pattern. Then, a novel framework for deep pre-training model is introduced to learn protein embeddings. After pre-training, our method can be easily fine-tuned for different protein tasks, including amino acid-level prediction task (e.g. secondary structure prediction), amino acid pair-level prediction task (e.g. contact prediction) and also protein-level prediction task (remote homology prediction, protein function prediction). Experiments show that our approach achieves significant improvements in all tasks and outperforms the previous methods. We also provide detailed ablation studies and analysis for our protein tokenizer and training framework.

Generalizable transcriptome-based tumor malignant level evaluation and molecular subtyping towards precision oncology.

In cancer treatment, therapeutic strategies that integrate tumor-specific characteristics (i.e., precision oncology) are widely implemented to provide clinical benefits for cancer patients. Here, through in-depth integration of tumor transcriptome and patients' prognoses across cancers, we investigated dysregulated and prognosis-associated genes and catalogued such important genes in a cancer type-dependent manner. Utilizing the expression matrices of these genes, we built models to quantitatively evaluate the malignant levels of tumors across cancers, which could add value to the clinical staging system for improved prediction of patients' survival. Furthermore, we performed a transcriptome-based molecular subtyping on hepatocellular carcinoma, which revealed three subtypes with significantly diversified clinical outcomes, mutation landscapes, immune microenvironment, and dysregulated pathways. As tumor transcriptome was commonly profiled in clinical practice with low experimental complexity and cost, this work proposed easy-to-perform approaches for practical clinical promotion towards better healthcare and precision oncology of cancer patients.

Sequential regioselective arylation of pyrazolones with diaryliodonium salts.

The introduction of aromatic substituents into organic compounds significantly alters their physical and chemical characteristics. Yet, achieving precise control over the site-selectivity of arylation continues to pose a considerable challenge. We present here a controllable method for the site-selective mono-, di-, and triarylation of pyrazolone with diaryliodonium salts. The method showcases robustness, flexibility, and excellent compatibility with a broad range of functional groups. It enables control over both the site of arylation and the number of aryl additions. Specifically, three of the four substitutable positions in pyrazolone can be selectively arylated, effectively producing four products under controlled conditions. Additionally, the method supports one-pot sequential arylation, leading to an array of products with diverse aromatic substituents. Control experiments revealed the specific conditions of each reaction step.

NIR triggered polydopamine coated cerium dioxide nanozyme for ameliorating acute lung injury via enhanced ROS scavenging.

Acute lung injury (ALI) is a life threatening disease in critically ill patients, and characterized by excessive reactive oxygen species (ROS) and inflammatory factors levels in the lung. Multiple evidences suggest that nanozyme with diversified catalytic capabilities plays a vital role in this fatal lung injury. At present, we developed a novel class of polydopamine (PDA) coated cerium dioxide (CeO2) nanozyme (Ce@P) that acts as the potent ROS scavenger for scavenging intracellular ROS and suppressing inflammatory responses against ALI. Herein, we aimed to identify that Ce@P combining with NIR irradiation could further strengthen its ROS scavenging capacity. Specifically, NIR triggered Ce@P exhibited the most potent antioxidant and anti-inflammatory behaviors in lipopolysaccharide (LPS) induced macrophages through decreasing the intracellular ROS levels, down-regulating the levels of TNF-α, IL-1ß and IL-6, up-regulating the level of antioxidant cytokine (SOD-2), inducing M2 directional polarization (CD206 up-regulation), and increasing the expression level of HSP70. Besides, we performed intravenous (IV) injection of Ce@P in LPS induced ALI rat model, and found that it significantly accumulated in the lung tissue for 6 h after injection. It was also observed that Ce@P + NIR presented the superior behaviors of decreasing lung inflammation, alleviating diffuse alveolar damage, as well as promoting lung tissue repair. All in all, it has developed the strategy of using Ce@P combining with NIR irradiation for the synergistic enhanced treatment of ALI, which can serve as a promising therapeutic strategy for the clinical treatment of ROS derived diseases as well.

Matrix Metalloproteinase and Aortic Aneurysm: A Two-sample Mendelian Randomization Study.

BACKGROUND: Studies have linked matrix metalloproteinases (MMPs) to both thoracic aortic aneurysm and abdominal aortic aneurysm (TAA and AAA). The precise MMPs entailed in this procedure, however, were still unknown. This study used a two-sample Mendelian randomization (MR) analysis to look into the causal relationship between MMPs and the risk of TAA and AAA. METHODS: Eight MMPs, including MMP-1, MMP-2, MMP-3, MMP-8, MMP-9, MMP-10, MMP-12, and MMP-13, were found among people of European ancestry with accessible Genome-Wide Association Studies (GWAS). We employed the findings from Genome-Wide Association Studies (GWAS) for 8 MMPs, and TAA and AAA from the FinnGen consortiums (3,201 cases and 317,899 controls, respectively) were used in a two-sample MR analysis. The primary method of analysis for MR was the inverse variance weighted (IVW) method, along with analyses of heterogeneity and horizontal pleiotropy. 31 single-nucleotide polymorphisms connected to MMP were retrieved. RESULTS: IVW demonstrated a negative causal association between TAA and AAA and serum MMP-12 levels. The incidence of TAA decreased by 1.031% for every 1 ng/mL increase in serum MMP-12 [odds ratio (OR) = 0.897, 95% confidence interval (CI): 0.831-0.968, P = 0.005]. The incidence of AAA fell by 1.653% (OR = 0.835, 95% CI: 0.752-0.926, P = 0.001) for every 1 ng/mL increase in serum MMP-12. There was no horizontal pleiotropy or heterogeneity in the MR data (P > 0.05). CONCLUSIONS: The levels of TAA and AAA and serum MMP-12 are causally related. MMP-12 is a factor that reduces the risk of AAA and TTA. Our study suggested that MMP-12 level is causally associated with a decreased risk of TAA and AAA.

Photodetector with a metalens packaging module for visible light communication based on RGBY illumination LED light source.

A novel, to the best of our knowledge, photodetector with a metalens packaging module used as the visible light communication (VLC) receiver is proposed and designed. An LED consisting of red, green, blue, and yellow chips (RGBY-LED) is adopted as the transmitter for intensity modulation direct detection VLC systems. A metalens array with a numerical aperture (NA) of 0.707 used as a polarization-insensitive planar lens of the VLC system receiver is designed at wavelengths of 457, 523, 592, and 623 nm corresponding to blue, green, yellow, and red for high efficiency. Compared with a traditional Fresnel lens positive-intrinsic-negative (PIN) photodetector module as the VLC receiver, the introduction of a metalens module can decrease the form factor of the VLC receiver module and, in particular, it is much thinner. The combination of the multi-color LED transmitter and photodetector metalens packaging module receiver can increase the modulation bandwidth due to four different wavelengths used for the VLC system. Finite-difference time domain (FDTD) simulations are performed to validate the performance of the photodetector with a metalens module. It is revealed that the corresponding efficiencies of 57.5%, 55.4%, 57%, and 56.3% were achieved at wavelengths of 623, 592, 523, and 457 nm, respectively, based on a metalens array with a 0.707 NA and 2.5 µm radius of the active area of the photodetector. It is a promising technology for indoor VLC systems such as those for smart phones and other Internet of Things devices due to the need for compact packaging for the receiver.

Pseudolaric acid B exerts an antifungal effect and targets SIRT1 to ameliorate inflammation by regulating Nrf2/NF-κB pathways in fungal keratitis.

Fungal keratitis (FK) is a vision-threatening infection. We aimed to explore the antifungal and anti-inflammatory effects of pseudolaric acid B (PAB) on FK and the underlying mechanisms involved. Network pharmacology utilized to acquire the potential target genes, and silent information regulator 1 (SIRT1) was consistently downregulated in Gene Expression Omnibus dataset and clinical samples. Molecular docking analysis showed that PAB and SIRT1 had good binding activity. No toxicity was observed in vivo and in vitro with a PAB concentration below 0.3 µM. PAB exerted its antifungal activity by destroying the integrity of hyphae, and alleviated the severity of FK in rats by decreasing clinical scores, fungal burden and inhibiting inflammatory cell infiltration. PAB increased SIRT1 to regulate the crosstalk between nuclear factor erythroid 2-related factor 2 (Nrf2) and nuclear factor kappa-B (NF-κB), decreasing the levels of inflammatory cytokines tumor necrosis factor (TNF)-α, interleukin (IL)-1ß, and IL-6; and pattern recognition receptors, C-type lectin domain containing 7A (Dectin-1), lectin-like oxidized low-density lipoprotein receptor 1 (LOX-1), toll like receptor (TLR)-2, and TLR4 both in vivo and in vitro. However, this anti-inflammatory effect of PAB was abolished by the SIRT1 inhibitor EX527. This study provides new evidence that PAB has antifungal and anti-inflammatory effects in FK and may provide a novel therapeutic strategy for the treatment of FK.

Optogenetic Spreading Depolarizations Do Not Worsen Acute Ischemic Stroke Outcome.

BACKGROUND: Spreading depolarizations (SDs) are believed to contribute to injury progression and worsen outcomes in focal cerebral ischemia because exogenously induced SDs have been associated with enlarged infarct volumes. However, previous studies used highly invasive methods to trigger SDs that can directly cause tissue injury (eg, topical KCl) and confound the interpretation. Here, we tested whether SDs indeed enlarge infarcts when induced via a novel, noninjurious method using optogenetics. METHODS: Using transgenic mice expressing channelrhodopsin-2 in neurons (Thy1-ChR2-YFP), we induced 8 optogenetic SDs to trigger SDs noninvasively at a remote cortical location in a noninjurious manner during 1-hour distal microvascular clip or proximal an endovascular filament occlusion of the middle cerebral artery. Laser speckle imaging was used to monitor cerebral blood flow. Infarct volumes were then quantified at 24 or 48 hours. RESULTS: Infarct volumes in the optogenetic SD arm did not differ from the control arm in either distal or proximal middle cerebral artery occlusion, despite a 6-fold and 4-fold higher number of SDs, respectively. Identical optogenetic illumination in wild-type mice did not affect the infarct volume. Full-field laser speckle imaging showed that optogenetic stimulation did not affect the perfusion in the peri-infarct cortex. CONCLUSIONS: Altogether, these data show that SDs induced noninvasively using optogenetics do not worsen tissue outcomes. Our findings compel a careful reexamination of the notion that SDs are causally linked to infarct expansion.

Spreading Depolarizations Suppress Hematoma Growth in Hyperacute Intracerebral Hemorrhage in Mice.

BACKGROUND: Spreading depolarizations (SDs) occur in all types of brain injury and may be associated with detrimental effects in ischemic stroke and subarachnoid hemorrhage. While rapid hematoma growth during intracerebral hemorrhage triggers SDs, their role in intracerebral hemorrhage is unknown. METHODS: We used intrinsic optical signal and laser speckle imaging, combined with electrocorticography, to investigate the effects of SD on hematoma growth during the hyperacute phase (0-4 hours) after intracortical collagenase injection in mice. Hematoma expansion, SDs, and cerebral blood flow were simultaneously monitored under normotensive and hypertensive conditions. RESULTS: Spontaneous SDs erupted from the vicinity of the hematoma during rapid hematoma growth. We found that hematoma growth slowed down by >60% immediately after an SD. This effect was even stronger in hypertensive animals with faster hematoma growth. To establish causation, we exogenously induced SDs (every 30 minutes) at a remote site by topical potassium chloride application and found reduced hematoma growth rate and final hemorrhage volume (18.2±5.8 versus 10.7±4.1 mm3). Analysis of cerebral blood flow using laser speckle flowmetry revealed that suppression of hematoma growth by spontaneous or induced SDs coincided and correlated with the characteristic oligemia in the wake of SD, implicating the vasoconstrictive effect of SD as one potential mechanism of action. CONCLUSIONS: Our findings reveal that SDs limit hematoma growth during the early hours of intracerebral hemorrhage and decrease final hematoma volume.

HER2-targeting antibody-drug conjugate RC48 alone or in combination with immunotherapy for locally advanced or metastatic urothelial carcinoma: a multicenter, real-world study.

BACKGROUND: Phase II trials showed the efficacy of anti-HER2 RC48-ADC (disitamab vedotin) for HER2-positive metastatic urothelial carcinoma (UC). This study evaluated RC48 alone verses in combination with immunotherapy for locally advanced or metastatic UC using real-world data. METHODS: This retrospective, multicenter, real-world study included patients with locally advanced or metastatic UC who received RC48 in five hospitals in China between July 2021 and April 2022. The outcomes were progression-free survival (PFS), overall survival (OS), objective response rate (ORR), disease control rate (DCR), and adverse events. RESULTS: Thirty-six patients were included. The patients were 47-87 years, and 26 (72.2%) were male. Eighteen patients received RC48 alone, and 18 received RC48 combined with a programmed death-1 antibody. The median PFS was 5.4 months. The median OS was not reached. The 6-month and 1-year PFS rates were 38.8% and 15.5%, respectively. The 1-year OS rate was 79.6%. Fourteen (38.9%) patients achieved a partial response, and the ORR was 38.9%. Eleven patients had stable disease, and the DCR was 69.4%. The median PFS for patients who received RC48 combined with immunotherapy and those who received RC48 alone was 8.5 and 5.4 months, respectively. The main treatment-related adverse events included anemia, hypoesthesia, fatigue, and elevated transaminase. No treatment-related death occurred. CONCLUSION: RC48 alone or combined with immunotherapy might benefit patients with locally advanced or metastatic UC, regardless of impaired renal function.

The genetic variation in drought resistance in eighteen perennial ryegrass varieties and the underlying adaptation mechanisms.

BACKGROUND: Drought resistance is a complex characteristic closely related to the severity and duration of stress. Perennial ryegrass (Lolium perenne L.) has no distinct drought tolerance but often encounters drought stress seasonally. Although the response of perennial ryegrass to either extreme or moderate drought stress has been investigated, a comprehensive understanding of perennial ryegrass response to both conditions of drought stress is currently lacking. RESULTS: In this study, we investigated the genetic variation in drought resistance in 18 perennial ryegrass varieties under both extreme and moderate drought conditions. The performance of these varieties exhibited obvious diversity, and the survival of perennial ryegrass under severe stress was not equal to good growth under moderate drought stress. 'Sopin', with superior performance under both stress conditions, was the best-performing variety. Transcriptome, physiological, and molecular analyses revealed that 'Sopin' adapted to drought stress through multiple sophisticated mechanisms. Under stress conditions, starch and sugar metabolic enzymes were highly expressed, while CslA was expressed at low levels in 'Sopin', promoting starch degradation and soluble sugar accumulation. The expression and activity of superoxide dismutase were significantly higher in 'Sopin', while the activity of peroxidase was lower, allowing for 'Sopin' to maintain a better balance between maintaining ROS signal transduction and alleviating oxidative damage. Furthermore, drought stress-related transcriptional and posttranscriptional regulatory mechanisms, including the upregulation of transcription factors, kinases, and E3 ubiquitin ligases, facilitate abscisic acid and stress signal transduction. CONCLUSION: Our study provides insights into the resistance of perennial ryegrass to both extreme and moderate droughts and the underlying mechanisms by which perennial ryegrass adapts to drought conditions.

Back translation for molecule generation.

MOTIVATION: Molecule generation, which is to generate new molecules, is an important problem in bioinformatics. Typical tasks include generating molecules with given properties, molecular property improvement (i.e. improving specific properties of an input molecule), retrosynthesis (i.e. predicting the molecules that can be used to synthesize a target molecule), etc. Recently, deep-learning-based methods received more attention for molecule generation. The labeled data of bioinformatics is usually costly to obtain, but there are millions of unlabeled molecules. Inspired by the success of sequence generation in natural language processing with unlabeled data, we would like to explore an effective way of using unlabeled molecules for molecule generation. RESULTS: We propose a new method, back translation for molecule generation, which is a simple yet effective semisupervised method. Let X be the source domain, which is the collection of properties, the molecules to be optimized, etc. Let Y be the target domain which is the collection of molecules. In particular, given a main task which is about to learn a mapping from the source domain X to the target domain Y, we first train a reversed model g for the Y to X mapping. After that, we use g to back translate the unlabeled data in Y to X and obtain more synthetic data. Finally, we combine the synthetic data with the labeled data and train a model for the main task. We conduct experiments on molecular property improvement and retrosynthesis, and we achieve state-of-the-art results on four molecule generation tasks and one retrosynthesis benchmark, USPTO-50k. AVAILABILITY AND IMPLEMENTATION: Our code and data are available at https://github.com/fyabc/BT4MolGen. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Discovering drug-target interaction knowledge from biomedical literature.

MOTIVATION: The interaction between drugs and targets (DTI) in human body plays a crucial role in biomedical science and applications. As millions of papers come out every year in the biomedical domain, automatically discovering DTI knowledge from biomedical literature, which are usually triplets about drugs, targets and their interaction, becomes an urgent demand in the industry. Existing methods of discovering biological knowledge are mainly extractive approaches that often require detailed annotations (e.g. all mentions of biological entities, relations between every two entity mentions, etc.). However, it is difficult and costly to obtain sufficient annotations due to the requirement of expert knowledge from biomedical domains. RESULTS: To overcome these difficulties, we explore an end-to-end solution for this task by using generative approaches. We regard the DTI triplets as a sequence and use a Transformer-based model to directly generate them without using the detailed annotations of entities and relations. Further, we propose a semi-supervised method, which leverages the aforementioned end-to-end model to filter unlabeled literature and label them. Experimental results show that our method significantly outperforms extractive baselines on DTI discovery. We also create a dataset, KD-DTI, to advance this task and release it to the community. AVAILABILITY AND IMPLEMENTATION: Our code and data are available at https://github.com/bert-nmt/BERT-DTI. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.