Search

Computers can already be programmed for superhuman pattern recognition of images and text. For machines to discover novel molecules, they must first be trained to sort through the many characteristics of molecules and determine which properties should be retained, suppressed, or enhanced to optimize functions of interest. Machines need to be able to understand, read, write, and eventually create new molecules. Today, this creative process relies on deep generative models, which have gained popularity since powerful deep neural networks were introduced to generative model frameworks. In recent years, they have demonstrated excellent ability to model complex distribution of real-word data (

Knowledge about the relations between biomedical entities (such as drugs and targets) is widely distributed in more than 30 million research articles and consistently plays an important role in the development of biomedical science. In this work, we propose a novel machine learning framework, named BERE, for automatically extracting biomedical relations from large-scale literature repositories. BERE uses a hybrid encoding network to better represent each sentence from both semantic and syntactic aspects, and employs a feature aggregation network to make predictions after considering all relevant statements. More importantly, BERE can also be trained without any human annotation via a distant supervision technique. Through extensive tests, BERE has demonstrated promising performance in extracting biomedical relations, and can also find meaningful relations that were not reported in existing databases, thus providing useful hints to guide wet-lab experiments and advance the biological knowledge discovery process. A lot of scientific literature is unstructured, which makes extracting information for biomedical databases difficult. Hong and colleagues show that a distant supervision approach, using latent tree learning and recurrent units, can extract drug–target interactions from literature that were previously unknown.

Motivation Quantitative structure–activity relationship (QSAR) and drug–target interaction (DTI) prediction are both commonly used in drug discovery. Collaboration among pharmaceutical institutions can lead to better performance in both QSAR and DTI prediction. However, the drug-related data privacy and intellectual property issues have become a noticeable hindrance for inter-institutional collaboration in drug discovery. Results We have developed two novel algorithms under secure multiparty computation (MPC), including QSARMPC and DTIMPC, which enable pharmaceutical institutions to achieve high-quality collaboration to advance drug discovery without divulging private drug-related information. QSARMPC, a neural network model under MPC, displays good scalability and performance and is feasible for privacy-preserving collaboration on large-scale QSAR prediction. DTIMPC integrates drug-related heterogeneous network data and accurately predicts novel DTIs, while keeping the drug information confidential. Under several experimental settings that reflect the situations in real drug discovery scenarios, we have demonstrated that DTIMPC possesses significant performance improvement over the baseline methods, generates novel DTI predictions with supporting evidence from the literature and shows the feasible scalability to handle growing DTI data. All these results indicate that QSARMPC and DTIMPC can provide practically useful tools for advancing privacy-preserving drug discovery. Availability and implementation The source codes of QSARMPC and DTIMPC are available on the GitHub: https://github.com/rongma6/QSARMPC_DTIMPC.git. Supplementary information Supplementary data are available at Bioinformatics online.

Summary The Columbia Cancer Target Discovery and Development (CTD2) Center is developing PANACEA, a resource comprising dose-responses and RNA sequencing (RNA-seq) profiles of 25 cell lines perturbed with ∼400 clinical oncology drugs, to study a tumor-specific drug mechanism of action. Here, this resource serves as the basis for a DREAM Challenge assessing the accuracy and sensitivity of computational algorithms for de novo drug polypharmacology predictions. Dose-response and perturbational profiles for 32 kinase inhibitors are provided to 21 teams who are blind to the identity of the compounds. The teams are asked to predict high-affinity binding targets of each compound among ∼1,300 targets cataloged in DrugBank. The best performing methods leverage gene expression profile similarity analysis as well as deep-learning methodologies trained on individual datasets. This study lays the foundation for future integrative analyses of pharmacogenomic data, reconciliation of polypharmacology effects in different tumor contexts, and insights into network-based assessments of drug mechanisms of action., Graphical abstract, Highlights • Drug-perturbed RNA sequencing data can be used to identify drug targets • Technology-based drug-target definitions often subsume literature definitions • Literature and screening datasets provide complementary information on drug mechanisms, Douglass et al. report the results of a crowdsourced challenge to develop machine-learning algorithms that use drug-perturbed transcriptome data to rapidly predict drug targets on a proteomic scale. Winning methods effectively predicted off-target binding of clinical kinase inhibitors and clarified disparate literature on these drugs’ mechanisms of action.

Peptide-protein interactions are involved in various fundamental cellular functions and their identification is crucial for designing efficacious peptide therapeutics. Recently, a number of computational methods have been developed to predict peptide-protein interactions. However, most of the existing prediction approaches heavily depend on high-resolution structure data. Here, we present a deep learning framework for multi-level peptide-protein interaction prediction, called CAMP, including binary peptide-protein interaction prediction and corresponding peptide binding residue identification. Comprehensive evaluation demonstrated that CAMP can successfully capture the binary interactions between peptides and proteins and identify the binding residues along the peptides involved in the interactions. In addition, CAMP outperformed other state-of-the-art methods on binary peptide-protein interaction prediction. CAMP can serve as a useful tool in peptide-protein interaction prediction and identification of important binding residues in the peptides, which can thus facilitate the peptide drug discovery process., Peptide-protein interactions play fundamental roles in cellular processes and are crucial for designing peptide therapeutics. Here, the authors present a deep learning framework for simultaneously predicting peptide-protein interactions and identifying peptide binding residues involved in the interactions.

Accurate identification of compound–protein interactions (CPIs) in silico may deepen our understanding of the underlying mechanisms of drug action and thus remarkably facilitate drug discovery and development. Conventional similarity- or docking-based computational methods for predicting CPIs rarely exploit latent features from currently available large-scale unlabeled compound and protein data and often limit their usage to relatively small-scale datasets. In the present study, we propose DeepCPI, a novel general and scalable computational framework that combines effective feature embedding (a technique of representation learning) with powerful deep learning methods to accurately predict CPIs at a large scale. DeepCPI automatically learns the implicit yet expressive low-dimensional features of compounds and proteins from a massive amount of unlabeled data. Evaluations of the measured CPIs in large-scale databases, such as ChEMBL and BindingDB, as well as of the known drug–target interactions from DrugBank, demonstrated the superior predictive performance of DeepCPI. Furthermore, several interactions among small-molecule compounds and three G protein-coupled receptor targets (glucagon-like peptide-1 receptor, glucagon receptor, and vasoactive intestinal peptide receptor) predicted using DeepCPI were experimentally validated. The present study suggests that DeepCPI is a useful and powerful tool for drug discovery and repositioning. The source code of DeepCPI can be downloaded from https://github.com/FangpingWan/DeepCPI. Keywords: Deep learning, Machine learning, Drug discovery, In silico drug screening, Compound–protein interaction prediction

Motivation Prediction of peptide binding to the major histocompatibility complex (MHC) plays a vital role in the development of therapeutic vaccines for the treatment of cancer. Algorithms with improved correlations between predicted and actual binding affinities are needed to increase precision and reduce the number of false positive predictions. Results We present ACME (Attention-based Convolutional neural networks for MHC Epitope binding prediction), a new pan-specific algorithm to accurately predict the binding affinities between peptides and MHC class I molecules, even for those new alleles that are not seen in the training data. Extensive tests have demonstrated that ACME can significantly outperform other state-of-the-art prediction methods with an increase of the Pearson correlation coefficient between predicted and measured binding affinities by up to 23 percentage points. In addition, its ability to identify strong-binding peptides has been experimentally validated. Moreover, by integrating the convolutional neural network with attention mechanism, ACME is able to extract interpretable patterns that can provide useful and detailed insights into the binding preferences between peptides and their MHC partners. All these results have demonstrated that ACME can provide a powerful and practically useful tool for the studies of peptide–MHC class I interactions. Availability and implementation ACME is available as an open source software at https://github.com/HYsxe/ACME. Supplementary information Supplementary data are available at Bioinformatics online.

RNA polymerase II (Pol II) generally pauses at certain positions along gene bodies, thereby interrupting the transcription elongation process, which is often coupled with various important biological functions, such as precursor mRNA splicing and gene expression regulation. Characterizing the transcriptional elongation dynamics can thus help us understand many essential biological processes in eukaryotic cells. However, experimentally measuring Pol II elongation rates is generally time and resource consuming. We developed PEPMAN (polymerase II elongation pausing modeling through attention-based deep neural network), a deep learning-based model that accurately predicts Pol II pausing sites based on the native elongating transcript sequencing (NET-seq) data. Through fully taking advantage of the attention mechanism, PEPMAN is able to decipher important sequence features underlying Pol II pausing. More importantly, we demonstrated that the analyses of the PEPMAN-predicted results around various types of alternative splicing sites can provide useful clues into understanding the cotranscriptional splicing events. In addition, associating the PEPMAN prediction results with different epigenetic features can help reveal important factors related to the transcription elongation process. All these results demonstrated that PEPMAN can provide a useful and effective tool for modeling transcription elongation and understanding the related biological factors from available high-throughput sequencing data.

Understanding the molecular properties (e.g., physical, chemical or physiological characteristics and biological activities) of small molecules plays essential roles in biomedical researches. The accumulating amount of datasets has enabled the development of data-driven computational methods, especially the machine learning based methods, to address the molecular property prediction tasks. Due to the high cost of obtaining experimental labels, the datasets of individual tasks generally contain limited amount of data, which inspired the application of transfer learning to boost the performance of the molecular property prediction tasks. Our analyses revealed that simultaneously considering similar tasks, rather than randomly chosen ones, can significantly improve the performance of transfer learning in this field. To provide accurate estimation of task similarity, we proposed an effective and interpretable computational tool, named Molecular Tasks Similarity Estimator (MoTSE). By extracting task-related local and global knowledge from pretrained graph neural networks (GNNs), MoTSE projects individual tasks into a latent space and measures the distance between the embedded vectors to derive the task similarity estimation and thus enhance the molecular prediction results. We have validated that the task similarity estimated by MoTSE can serve as a useful guidance to design a more accurate transfer learning strategy for molecular property prediction. Experimental results showed that such a strategy greatly outperformed baseline methods including training from scratch and multitask learning. Moreover, MoTSE can provide interpretability for the estimated task similarity, through visualizing the important loci in the molecules attributed by the attribution method employed in MoTSE. In summary, MoTSE can provide an accurate method for estimating the molecular property task similarity for effective transfer learning, with good interpretability for the learned chemical or biological insights underlying the intrinsic principles of the task similarity.

The Columbia Cancer Target Discovery and Development (CTD2) Center has developed PANACEA (PANcancer Analysis of Chemical Entity Activity), a collection of dose-response curves and perturbational profiles for 400 clinical oncology drugs in cell lines selected to optimally represent 19 cancer subtypes. This resource, developed to study tumor-specific drug mechanism of action, was instrumental in hosting a DREAM Challenge to assess computational models for de novo drug polypharmacology prediction. Dose-response and perturbational profiles for 32 kinase inhibitors were provided to 21 participating teams who were asked to predict high-affinity binding target among 255 possible protein kinases. Best performing methods leveraged both gene expression profile similarity analysis, and deep-learning methodologies trained on individual datasets. This study lays the foundation for future integrative analyses of pharmacogenomic data, reconciliation of polypharmacology effects in different tumor contexts, and insights into network-based assessment of context-specific drug mechanism of action.

SUMMARYThe Columbia Cancer Target Discovery and Development (CTD2) Center has developed PANACEA (PANcancer Analysis of Chemical Entity Activity), a collection of dose-response curves and perturbational profiles for 400 clinical oncology drugs in cell lines selected to optimally represent 19 cancer subtypes. This resource, developed to study tumor-specific drug mechanism of action, was instrumental in hosting a DREAM Challenge to assess computational models for de novo drug polypharmacology prediction. Dose-response and perturbational profiles for 32 kinase inhibitors were provided to 21 participating teams, who did not know the identity or nature of the compounds, and they were asked to predict high-affinity binding among ~1,300 possible protein targets. Best performing methods leveraged both gene expression profile similarity analysis, and deep-learning methodologies trained on individual datasets. This study lays the foundation for future integrative analyses of pharmacogenomic data, reconciliation of polypharmacology effects in different tumor contexts, and insights into network-based assessment of context-specific drug mechanism of action.

Peptide-protein interactions (PepPIs) are involved in various fundamental cellular functions and their identification is crucial for designing efficacious peptide therapeutics. To facilitate the peptide drug discovery process, a number of computational methods have been developed to predict peptide-protein interactions. However, most of the existing prediction approaches heavily depend on high-resolution structure data. Although several deep-learning-based frameworks have been proposed to predict compound-protein interactions or protein-protein interactions, few of them are particularly designed to specifically predict peptide-protein interactions. In this paper, We present a sequence-basedConvolutionalAttention-based neural network forMultifaceted prediction ofPeptide-protein interactions, calledCAMP, including predicting binary peptide-protein interactions and corresponding binding residues in the peptides. We also construct a benchmark dataset containing high-quality peptide-protein interaction pairs with the corresponding peptide binding residues for model training and evaluation. CAMP incorporates convolution neural network architectures and attention mechanism to fully exploit informative sequence-based features, including secondary structures, physicochemical properties, intrinsic disorder features and position-specific scoring matrix of the protein. Systematical evaluation of our benchmark dataset demonstrates that CAMP outperforms the state-of-the-art baseline methods on binary peptide-protein interaction prediction. In addition, CAMP can successfully identify the binding residues involved non-covalent interactions for peptides. These results indicate that CAMP can serve as a useful tool in peptide-protein interaction prediction and peptide binding site identification, which can thus greatly facilitate the peptide drug discovery process. The source code of CAMP can be found inhttps://github.com/twopin/CAMP.

The global spread of SARS-CoV-2 requires an urgent need to find effective therapeutics for the treatment of COVID-19. We developed a data-driven drug repositioning framework, which applies both machine learning and statistical analysis approaches to systematically integrate and mine large-scale knowledge graph, literature and transcriptome data to discover the potential drug candidates against SARS-CoV-2. The retrospective study using the past SARS-CoV and MERS-CoV data demonstrated that our machine learning based method can successfully predict effective drug candidates against a specific coronavirus. Ourin silicoscreening followed by wet-lab validation indicated that a poly-ADP-ribose polymerase 1 (PARP1) inhibitor, CVL218, currently in Phase I clinical trial, may be repurposed to treat COVID-19. Ourin vitroassays revealed that CVL218 can exhibit effective inhibitory activity against SARS-CoV-2 replication without obvious cytopathic effect. In addition, we showed that CVL218 is able to suppress the CpG-induced IL-6 production in peripheral blood mononuclear cells, suggesting that it may also have anti-inflammatory effect that is highly relevant to the prevention immunopathology induced by SARS-CoV-2 infection. Further pharmacokinetic and toxicokinetic evaluation in rats and monkeys showed a high concentration of CVL218 in lung and observed no apparent signs of toxicity, indicating the appealing potential of this drug for the treatment of the pneumonia caused by SARS-CoV-2 infection. Moreover, molecular docking simulation suggested that CVL218 may bind to the N-terminal domain of nucleocapsid (N) protein of SARS-CoV-2, providing a possible model to explain its antiviral action. We also proposed several possible mechanisms to explain the antiviral activities of PARP1 inhibitors against SARS-CoV-2, based on the data present in this study and previous evidences reported in the literature. In summary, the PARP1 inhibitor CVL218 discovered by our data-driven drug repositioning framework can serve as a potential therapeutic agent for the treatment of COVID-19.

Despite decades of intensive search for compounds that modulate the activity of particular targets, there are currently small-molecules available only for a small proportion of the human proteome. Effective approaches are therefore required to map the massive space of unexplored compound-target interactions for novel and potent activities. Here, we carried out a crowdsourced benchmarking of predictive models for kinase inhibitor potencies across multiple kinase families using unpublished bioactivity data. The top-performing predictions were based on kernel learning, gradient boosting and deep learning, and their ensemble resulted in predictive accuracy exceeding that of kinase activity assays. We then made new experiments based on the model predictions, which further improved the accuracy of experimental mapping efforts and identified unexpected potencies even for under-studied kinases. The open-source algorithms together with the novel bioactivities between 95 compounds and 295 kinases provide a resource for benchmarking new prediction algorithms and for extending the druggable kinome.

Computational approaches for inferring the mechanisms of compound-protein interactions (CPIs) can greatly facilitate drug development. Recently, although a number of deep learning based methods have been proposed to predict binding affinities and attempt to capture local interaction sites in compounds and proteins through neural attentions, they still lack a systematic evaluation on the interpretability of the identified local features. In addition, in these previous approaches, the exact matchings between interaction sites from compounds and proteins, which are generally important for understanding drug mechanisms of action, still remain unknown. Here, we compiled the first benchmark dataset containing the inter-molecular non-covalent interactions for more than 10,000 compound-protein pairs, and used it to systematically evaluate the interpretability of neural attentions in existing prediction models. We developed a multi-objective neural network, called MONN, to predict both non-covalent interactions and binding affinity for a given compound-protein pair. MONN uses convolution neural networks on molecular graphs of compounds and primary sequences of proteins to effectively capture the intrinsic features from both inputs, and also takes advantage of the predicted non-covalent interactions to further boost the accuracy of binding affinity prediction. Comprehensive evaluation demonstrated that while the previous neural attention based approaches fail to exhibit satisfactory interpretability results without extra supervision, MONN can successfully predict non-covalent interactions on our benchmark dataset as well as another independent dataset derived from the Protein Data Bank (PDB). Moreover, MONN can outperform other state-of-the-art methods in predicting compound-protein binding affinities. In addition, the pairwise interactions predicted by MONN displayed compatible and accordant patterns in chemical properties, which provided another evidence to support the strong predictive power of MONN. These results suggested that MONN can offer a powerful tool in predicting binding affinities of compound-protein pairs and also provide useful insights into understanding the molecular mechanisms of compound-protein interactions, which thus can greatly advance the drug discovery process. The source code of the MONN model and the dataset creation process can be downloaded from https://github.com/lishuya17/MONN.

Summary Computational approaches for understanding compound-protein interactions (CPIs) can greatly facilitate drug development. Recently, a number of deep-learning-based methods have been proposed to predict binding affinities and attempt to capture local interaction sites in compounds and proteins through neural attentions (i.e., neural network architectures that enable the interpretation of feature importance). Here, we compiled a benchmark dataset containing the inter-molecular non-covalent interactions for more than 10,000 compound-protein pairs and systematically evaluated the interpretability of neural attentions in existing models. We also developed a multi-objective neural network, called MONN, to predict both non-covalent interactions and binding affinities between compounds and proteins. Comprehensive evaluation demonstrated that MONN can successfully predict the non-covalent interactions between compounds and proteins that cannot be effectively captured by neural attentions in previous prediction methods. Moreover, MONN outperforms other state-of-the-art methods in predicting binding affinities. Source code for MONN is freely available for download at https://github.com/lishuya17/MONN .

MotivationAccurately predicting drug-target interactions (DTIs) in silico can guide the drug discovery process and thus facilitate drug development. Computational approaches for DTI prediction that adopt the systems biology perspective generally exploit the rationale that the properties of drugs and targets can be characterized by their functional roles in biological networks.ResultsInspired by recent advance of information passing and aggregation techniques that generalize the convolution neural networks (CNNs) to mine large-scale graph data and greatly improve the performance of many network-related prediction tasks, we develop a new nonlinear end-to-end learning model, called NeoDTI, that integrates diverse information from heterogeneous network data and automatically learns topology-preserving representations of drugs and targets to facilitate DTI prediction. The substantial prediction performance improvement over other state-of-the-art DTI prediction methods as well as several novel predicted DTIs with evidence supports from previous studies have demonstrated the superior predictive power of NeoDTI. In addition, NeoDTI is robust against a wide range of choices of hyperparameters and is ready to integrate more drug and target related information (e.g., compound-protein binding affinity data). All these results suggest that NeoDTI can offer a powerful and robust tool for drug development and drug repositioning.Availability and implementationThe source code and data used in NeoDTI are available at: https://github.com/FangpingWan/NeoDTI.Contactzengjy321@tsinghua.edu.cnSupplementary informationSupplementary data are available at Bioinformatics online.

Accurately identifying compound-protein interactions in silico can deepen our understanding of the mechanisms of drug action and significantly facilitate the drug discovery and development process. Traditional similarity-based computational models for compound-protein interaction prediction rarely exploit the latent features from current available large-scale unlabelled compound and protein data, and often limit their usage on relatively small-scale datasets. We propose a new scheme that combines feature embedding (a technique of representation learning) with deep learning for predicting compound-protein interactions. Our method automatically learns the low-dimensional implicit but expressive features for compounds and proteins from the massive amount of unlabelled data. Combining effective feature embedding with powerful deep learning techniques, our method provides a general computational pipeline for accurate compound-protein interaction prediction, even when the interaction knowledge of compounds and proteins is entirely unknown. Evaluations on current large-scale databases of the measured compound-protein affinities, such as ChEMBL and BindingDB, as well as known drug-target interactions from DrugBank have demonstrated the superior prediction performance of our method, and suggested that it can offer a useful tool for drug development and drug repositioning.