Predicting drug-target interaction based on sequence and structure information

Positive and unlabeled learning (PU learning) is a significant binary classification task in machine learning; it focuses on training accurate classifiers using positive data and unlabeled data. Most of the works in this area are based on a two-step strategy: the first step is to identify reliable negative examples from unlabeled examples, and the second step is to construct the classifiers based on the positive examples and the identified reliable negative examples using supervised learning methods. However, these methods always underutilize the remaining unlabeled data, which limits the performance of PU learning. Furthermore, many methods require the iterative solution of the formulated quadratic programming problems to obtain the final classifier, resulting in a large computational cost. In this paper, we propose a new method called the absolute value inequality support vector machine, which applies the concept of eccentricity to select reliable negative examples from unlabeled data and then constructs a classifier based on the positive examples, the selected negative examples, and the remaining unlabeled data. In addition, we apply a hyperparameter optimization technique to automatically search and select the optimal parameter values in the proposed algorithm. Numerical experimental results on ten real-world datasets demonstrate that our method is better than the other three benchmark algorithms.

Drug–target interaction prediction based on improved heterogeneous graph representation learning and feature projection classification

文本分类是信息检索的关键问题之一.提取更多的可信反例和构造准确高效的分类器是PU(positive andunlabeled)文本分类的两个重要问题.然而,在现有的可信反例提取方法中,很多方法提取的可信反例数量较少,构建的分类器质量有待提高.分别针对这两个重要步骤提供了一种基于聚类的半监督主动分类方法.与传统的反例提取方法不同,利用聚类技术和正例文档应与反例文档共享尽可能少的特征项这一特点,从未标识数据集中尽可能多地移除正例,从而可以获得更多的可信反例.结合SVM 主动学习和改进的Rocchio 构建分类器,并采用改进的TFIDF(term frequency inverse document frequency)进行特征提取,可以显著提高分类的准确度.分别在3 个不同的数据集中测试了分类结果(RCV1,Reuters-21578,20 Newsgoups).实验结果表明,基于聚类寻找可信反例可以在保持较低错误率的情况下获取更多的可信反例,而且主动学习方法的引入也显著提升了分类精度.

Deceptive reviews detection has attracted significant attention from both business and research communities. However, due to the difficulty of human labeling needed for supervised learning, the problem remains to be highly challenging. This paper proposed a novel angle to the problem by modeling PU (positive unlabeled) learning. A semi-supervised model, called mixing population and individual property PU learning (MPIPUL), is proposed. Firstly, some reliable negative examples are identified from the unlabeled dataset. Secondly, some representative positive examples and negative examples are generated based on LDA (Latent Dirichlet Allocation). Thirdly, for the remaining unlabeled examples (we call them spy examples), which can not be explicitly identified as positive and negative, two similarity weights are assigned, by which the probability of a spy example belonging to the positive class and the negative class are displayed. Finally, spy examples and their similarity weights are incorporated into SVM (Support Vector Machine) to build an accurate classifier. Experiments on gold-standard dataset demonstrate the effectiveness of MPIPUL which outperforms the state-of-the-art baselines.

Multi-domain spoken dialog system should be able to detect more than one domain from a user’s utterance. However, it is difficult to train an accurate binary classifier of a domain based on only positive and unlabeled examples. This paper improves hierarchical clustering algorithm to automatically identify reliable negative examples among unlabeled examples. This paper also verifies three linkage criteria that measure the distance between two clusters. In experiments, the proposed method resulted in the highest gain of F1 score compared to the existing methods.

BackgroundPredicting novel drug–target associations is important not only for developing new drugs, but also for furthering biological knowledge by understanding how drugs work and their modes of action. As more data about drugs, targets, and their interactions becomes available, computational approaches have become an indispensible part of drug target association discovery. In this paper we apply random walk with restart (RWR) method to a heterogeneous network of drugs and targets compiled from DrugBank database and investigate the performance of the method under parameter variation and choice of chemical fingerprint methods.ResultsWe show that choice of chemical fingerprint does not affect the performance of the method when the parameters are tuned to optimal values. We use a subset of the ChEMBL15 dataset that contains 2,763 associations between 544 drugs and 467 target proteins to evaluate our method, and we extracted datasets of bioactivity ≤1 and ≤10 μM activity cutoff. For 1 μM bioactivity cutoff, we find that our method can correctly predict nearly 47, 55, 60% of the given drug–target interactions in the test dataset having more than 0, 1, 2 drug target relations for ChEMBL 1 μM dataset in top 50 rank positions. For 10 μM bioactivity cutoff, we find that our method can correctly predict nearly 32.4, 34.8, 35.3% of the given drug–target interactions in the test dataset having more than 0, 1, 2 drug target relations for ChEMBL 1 μM dataset in top 50 rank positions. We further examine the associations between 110 popular top selling drugs in 2012 and 3,519 targets and find the top ten targets for each drug.ConclusionsWe demonstrate the effectiveness and promise of the approach—RWR on heterogeneous networks using chemical features—for identifying novel drug target interactions and investigate the performance.Electronic supplementary materialThe online version of this article (doi:10.1186/s13321-015-0089-z) contains supplementary material, which is available to authorized users.

Two birds with one stone: Classifying positive and unlabeled examples on uncertain data streams

Identification of interaction between drugs and target proteins plays an important role in discovering new drug candidates. However, through the experimental method to identify the drug-target interactions remain to be extremely time-consuming, expensive and challenging even nowadays. Therefore, it is urgent to develop new computational methods to predict potential drugtarget interactions (DTI). In this article, a novel computational model is developed for predicting potential drug-target interactions under the theory that each drug-target interaction pair can be represented by the structural properties from drugs and evolutionary information derived from proteins. Specifically, the protein sequences are encoded as Position-Specific Scoring Matrix (PSSM) descriptor which contains information of biological evolutionary and the drug molecules are encoded as fingerprint feature vector which represents the existence of certain functional groups or fragments. Four benchmark datasets involving enzymes, ion channels, GPCRs and nuclear receptors, are independently used for establishing predictive models with Rotation Forest (RF) model. The proposed method achieved the prediction accuracy of 91.3%, 89.1%, 84.1% and 71.1% for four datasets respectively. In order to make our method more persuasive, we compared our classifier with the state-of-theart Support Vector Machine (SVM) classifier. We also compared the proposed method with other excellent methods. Experimental results demonstrate that the proposed method is effective in the prediction of DTI, and can provide assistance for new drug research and development.

Existing deep learning methods have shown outstanding performance in predicting drug-target interactions. However, they still have limitations: (1) the over-reliance on locally extracted features by some single encoders, with insufficient consideration of global features, and (2) the inadequate modeling and learning of local crucial interaction sites in drug-target interaction pairs. In this study, we propose a novel drug-target interaction prediction model called the Neural Fingerprint and Self-Attention Mechanism (NFSA-DTI), which effectively integrates the local information of drug molecules and target sequences with their respective global features. The neural fingerprint method is used in this model to extract global features of drug molecules, while the self-attention mechanism is utilized to enhance CNN's capability in capturing the long-distance dependencies between the subsequences in the target amino acid sequence. In the feature fusion module, we improve the bilinear attention network by incorporating attention pooling, which enhances the model's ability to learn local crucial interaction sites in the drug-target pair. The experimental results on three benchmark datasets demonstrated that NFSA-DTI outperformed all baseline models in predictive performance. Furthermore, case studies illustrated that our model could provide valuable insights for drug discovery. Moreover, our model offers molecular-level interpretations.

A bagging SVM to learn from positive and unlabeled examples

Identification of drug target proteins is an important process in drug discovery and development. To perform predictions of large-scale drug-target interactions, many computational methods have been proposed using the information of target-target and drug-drug similarities into a network. In this work, we propose an enhanced heterogeneous network model with various biological aspects of target-target and drug-drug similarities to predict drug-target associations. Network propagation was performed through the heterogeneous network to predict new potential targets of a certain drug. Cross-validation and the Area Under the Receiver Operating Characteristic curve (AUROC) were then used to qualify the best model in the drug-target association prediction. The result shows that the model using the similarity measures based on drug-disease associations and target protein sequences significantly outperforms the other models and the original model. Overall, our best model yields an accuracy of 93.5%. In conclusion, the heterogeneous model with optimal similarity measures is a well-suited approach to predict a new target for a certain drug.

Identifying drug–target interactions (DTIs) is important for drug discovery. However, searching all drug–target spaces poses a major bottleneck. Therefore, recently many deep learning models have been proposed to address this problem. However, the developers of these deep learning models have neglected interpretability in model construction, which is closely related to a model’s performance. We hypothesized that training a model to predict important regions on a protein sequence would increase DTI prediction performance and provide a more interpretable model. Consequently, we constructed a deep learning model, named Highlights on Target Sequences (HoTS), which predicts binding regions (BRs) between a protein sequence and a drug ligand, as well as DTIs between them. To train the model, we collected complexes of protein–ligand interactions and protein sequences of binding sites and pretrained the model to predict BRs for a given protein sequence–ligand pair via object detection employing transformers. After pretraining the BR prediction, we trained the model to predict DTIs from a compound token designed to assign attention to BRs. We confirmed that training the BRs prediction model indeed improved the DTI prediction performance. The proposed HoTS model showed good performance in BR prediction on independent test datasets even though it does not use 3D structure information in its prediction. Furthermore, the HoTS model achieved the best performance in DTI prediction on test datasets. Additional analysis confirmed the appropriate attention for BRs and the importance of transformers in BR and DTI prediction. The source code is available on GitHub (https://github.com/GIST-CSBL/HoTS).

BackgroundOne of the most successful approaches to develop new small molecule therapeutics has been to start from a validated druggable protein target. However, only a small subset of potentially druggable targets has attracted significant research and development resources. The Illuminating the Druggable Genome (IDG) project develops resources to catalyze the development of likely targetable, yet currently understudied prospective drug targets. A central component of the IDG program is a comprehensive knowledge resource of the druggable genome.ResultsAs part of that effort, we have developed a framework to integrate, navigate, and analyze drug discovery data based on formalized and standardized classifications and annotations of druggable protein targets, the Drug Target Ontology (DTO). DTO was constructed by extensive curation and consolidation of various resources. DTO classifies the four major drug target protein families, GPCRs, kinases, ion channels and nuclear receptors, based on phylogenecity, function, target development level, disease association, tissue expression, chemical ligand and substrate characteristics, and target-family specific characteristics. The formal ontology was built using a new software tool to auto-generate most axioms from a database while supporting manual knowledge acquisition. A modular, hierarchical implementation facilitate ontology development and maintenance and makes use of various external ontologies, thus integrating the DTO into the ecosystem of biomedical ontologies. As a formal OWL-DL ontology, DTO contains asserted and inferred axioms. Modeling data from the Library of Integrated Network-based Cellular Signatures (LINCS) program illustrates the potential of DTO for contextual data integration and nuanced definition of important drug target characteristics. DTO has been implemented in the IDG user interface Portal, Pharos and the TIN-X explorer of protein target disease relationships.ConclusionsDTO was built based on the need for a formal semantic model for druggable targets including various related information such as protein, gene, protein domain, protein structure, binding site, small molecule drug, mechanism of action, protein tissue localization, disease association, and many other types of information. DTO will further facilitate the otherwise challenging integration and formal linking to biological assays, phenotypes, disease models, drug poly-pharmacology, binding kinetics and many other processes, functions and qualities that are at the core of drug discovery. The first version of DTO is publically available via the website http://drugtargetontology.org/, Github (http://github.com/DrugTargetOntology/DTO), and the NCBO Bioportal (http://bioportal.bioontology.org/ontologies/DTO). The long-term goal of DTO is to provide such an integrative framework and to populate the ontology with this information as a community resource.

Computational prediction of drug-target interactions (DTIs) and drug repositioning provides a low-cost and high-efficiency approach for drug discovery and development. The traditional social network-derived methods based on the naïve DTI topology information cannot predict potential targets for new chemical entities or failed drugs in clinical trials. There are currently millions of commercially available molecules with biologically relevant representations in chemical databases. It is urgent to develop novel computational approaches to predict targets for new chemical entities and failed drugs on a large scale. In this study, we developed a useful tool, namely substructure-drug-target network-based inference (SDTNBI), to prioritize potential targets for old drugs, failed drugs and new chemical entities. SDTNBI incorporates network and chemoinformatics to bridge the gap between new chemical entities and known DTI network. High performance was yielded in 10-fold and leave-one-out cross validations using four benchmark data sets, covering G protein-coupled receptors, kinases, ion channels and nuclear receptors. Furthermore, the highest areas under the receiver operating characteristic curve were 0.797 and 0.863 for two external validation sets, respectively. Finally, we identified thousands of new potential DTIs via implementing SDTNBI on a global network. As a proof-of-principle, we showcased the use of SDTNBI to identify novel anticancer indications for nonsteroidal anti-inflammatory drugs by inhibiting AKR1C3, CA9 or CA12. In summary, SDTNBI is a powerful network-based approach that predicts potential targets for new chemical entities on a large scale and will provide a new tool for DTI prediction and drug repositioning. The program and predicted DTIs are available on request.

The problem of PU Learning, i.e., learning classifiers with positive and unlabelled examples (but not negative examples), is very important in information retrieval and data mining. We address this problem through a novel approach: reducing it to the problem of learning classifiers for some meaningful multivariate performance measures. In particular, we show how a powerful machine learning algorithm, support vector machine, can be adapted to solve this problem. The effectiveness and efficiency of the proposed approach have been confirmed by our experiments on three real-world datasets.