GESSE: Predicting Drug Side Effects from Drug–Target Relationships

Non-Negative matrix factorization combined with kernel regression for the prediction of adverse drug reaction profiles

Motivation

Post-market unexpected Adverse Drug Reactions (ADRs) are associated with significant costs, in both financial burden and human health. Due to the high cost and time required to run clinical trials, there is significant interest in accurate computational methods that can aid in the prediction of ADRs for new drugs. As a machine learning task, ADR prediction is made more challenging due to a high degree of class imbalance and existing methods do not successfully balance the requirement to detect the minority cases (true positives for ADR), as measured by the Area Under the Precision-Recall (AUPR) curve with the ability to separate true positives from true negatives [as measured by the Area Under the Receiver Operating Characteristic (AUROC) curve]. Surprisingly, the performance of most existing methods is worse than a naïve method that attributes ADRs to drugs according to the frequency with which the ADR has been observed over all other drugs. The existing advanced methods applied do not lead to substantial gains in predictive performance.

Results

We designed a rigorous evaluation to provide an unbiased estimate of the performance of ADR prediction methods: Nested Cross-Validation and a hold-out set were adopted. Among the existing methods, Kernel Regression (KR) performed best in AUPR but had a disadvantage in AUROC, relative to other methods, including the naïve method. We proposed a novel method that combines non-negative matrix factorization with kernel regression, called VKR. This novel approach matched or exceeded the performance of existing methods, overcoming the weakness of the existing methods.

Availability

Code and data are available on https://github.com/YezhaoZhong/VKR.

Prediction of adverse drug reactions based on knowledge graph embedding

BACKGROUND

Adverse drug reactions (ADRs) are an important concern in the medication process and can pose a substantial economic burden for patients and hospitals. Because of the limitations of clinical trials, it is difficult to identify all possible ADRs of a drug before it is marketed. We developed a new model based on data mining technology to predict potential ADRs based on available drug data.

METHOD

Based on the Word2Vec model in Nature Language Processing, we propose a new knowledge graph embedding method that embeds drugs and ADRs into their respective vectors and builds a logistic regression classification model to predict whether a given drug will have ADRs.

RESULT

First, a new knowledge graph embedding method was proposed, and comparison with similar studies showed that our model not only had high prediction accuracy but also was simpler in model structure. In our experiments, the AUC of the classification model reached a maximum of 0.87, and the mean AUC was 0.863.

CONCLUSION

In this paper, we introduce a new method to embed knowledge graph to vectorize drugs and ADRs, then use a logistic regression classification model to predict whether there is a causal relationship between them. The experiment showed that the use of knowledge graph embedding can effectively encode drugs and ADRs. And the proposed ADRs prediction system is also very effective.

Adverse Drug Reaction Discovery Using a Tumor-Biomarker Knowledge Graph

Adverse drug reactions (ADRs) are a major public health concern, and early detection is crucial for drug development and patient safety. Together with the increasing availability of large-scale literature data, machine learning has the potential to predict unknown ADRs from current knowledge. By the machine learning methods, we constructed a Tumor-Biomarker Knowledge Graph (TBKG) which contains four types of node: Tumor, Biomarker, Drug, and ADR using biomedical literatures. Based on this knowledge graph, we not only discovered potential ADRs of antitumor drugs but also provided explanations. Experiments on real-world data show that our model can achieve 0.81 accuracy of three cross-validation and the ADRs discovery of Osimertinib was chosen for the clinical validation. Calculated ADRs of Osimertinib by our model consisted of the known ADRs which were in line with the official manual and some unreported rare ADRs in clinical cases. Results also showed that our model outperformed traditional co-occurrence methods. Moreover, each calculated ADRs were attached with the corresponding paths of “tumor-biomarker-drug” in the knowledge graph which could help to obtain in-depth insights into the underlying mechanisms. In conclusion, the tumor-biomarker knowledge-graph based approach is an explainable method for potential ADRs discovery based on biomarkers and might be valuable to the community working on the emerging field of biomedical literature mining and provide impetus for the mechanism research of ADRs.

Prediction of Adverse Drug Reactions by Combining Biomedical Tripartite Network and Graph Representation Model

As one of the primary contributors to high clinical attrition rates of drugs, toxicity evaluation is of critical significance to new drug discovery. Unsurprisingly, a vast number of computational methods have been developed at various stages of development pipeline to evaluate potential adverse drug reactions (ADRs). Despite previous success of these methods on individual ADR or certain drug family, there are great challenges to toxicity evaluation. In this study, a novel strategy was developed to predict the drug-ADR associations by combining deep learning and the biomedical tripartite network. This heterogeneous network contains biomedical linked data of three entities, for example, drugs, targets, and ADRs. For the first time, GraRep, a deep learning method for distributed representations, is introduced to learn graph representations and identify hidden features from the tripartite network which are further used for ADR prediction. Through this approach, drug-ADR associations could possibly be discovered from a systemic perspective. The accuracy of our method is 0.95 based on internal resource validation and 0.88 based on external resource validation. Moreover, our results show the prediction accuracy using the tripartite network is better than the one with bipartite network, suggesting the model performance can be improved with further enrichment on information. According to the result of 10-fold cross validation, the deep learning model outperforms two traditional methods (topology-based measures and chemical structure-based measures). Additionally, predictive models are also constructed using other deep learning methods, and comparable results are achieved. In summary, the biomedical tripartite network-based deep learning model proposed here proves to offer a promising solution for prediction of ADRs.

PyBioMed: a python library for various molecular representations of chemicals, proteins and DNAs and their interactions

Background

With the increasing development of biotechnology and informatics technology, publicly available data in chemistry and biology are undergoing explosive growth. Such wealthy information in these data needs to be extracted and transformed to useful knowledge by various data mining methods. Considering the amazing rate at which data are accumulated in chemistry and biology fields, new tools that process and interpret large and complex interaction data are increasingly important. So far, there are no suitable toolkits that can effectively link the chemical and biological space in view of molecular representation. To further explore these complex data, an integrated toolkit for various molecular representation is urgently needed which could be easily integrated with data mining algorithms to start a full data analysis pipeline.

Results

Herein, the python library PyBioMed is presented, which comprises functionalities for online download for various molecular objects by providing different IDs, the pretreatment of molecular structures, the computation of various molecular descriptors for chemicals, proteins, DNAs and their interactions. PyBioMed is a feature-rich and highly customized python library used for the characterization of various complex chemical and biological molecules and interaction samples. The current version of PyBioMed could calculate 775 chemical descriptors and 19 kinds of chemical fingerprints, 9920 protein descriptors based on protein sequences, more than 6000 DNA descriptors from nucleotide sequences, and interaction descriptors from pairwise samples using three different combining strategies. Several examples and five real-life applications were provided to clearly guide the users how to use PyBioMed as an integral part of data analysis projects. By using PyBioMed, users are able to start a full pipelining from getting molecular data, pretreating molecules, molecular representation to constructing machine learning models conveniently.

Conclusion

PyBioMed provides various user-friendly and highly customized APIs to calculate various features of biological molecules and complex interaction samples conveniently, which aims at building integrated analysis pipelines from data acquisition, data checking, and descriptor calculation to modeling. PyBioMed is freely available at http://projects.scbdd.com/pybiomed.html.

Knowledge graph prediction of unknown adverse drug reactions and validation in electronic health records

Unknown adverse reactions to drugs available on the market present a significant health risk and limit accurate judgement of the cost/benefit trade-off for medications. Machine learning has the potential to predict unknown adverse reactions from current knowledge. We constructed a knowledge graph containing four types of node: drugs, protein targets, indications and adverse reactions. Using this graph, we developed a machine learning algorithm based on a simple enrichment test and first demonstrated this method performs extremely well at classifying known causes of adverse reactions (AUC 0.92). A cross validation scheme in which 10% of drug-adverse reaction edges were systematically deleted per fold showed that the method correctly predicts 68% of the deleted edges on average. Next, a subset of adverse reactions that could be reliably detected in anonymised electronic health records from South London and Maudsley NHS Foundation Trust were used to validate predictions from the model that are not currently known in public databases. High-confidence predictions were validated in electronic records significantly more frequently than random models, and outperformed standard methods (logistic regression, decision trees and support vector machines). This approach has the potential to improve patient safety by predicting adverse reactions that were not observed during randomised trials.

A simple mathematical approach to the analysis of polypharmacology and polyspecificity data

There many possible types of drug-target interactions, because there are a surprising number of ways in which drugs and their targets can associate with one another.  These relationships are expressed as polypharmacology and polyspecificity.  Polypharmacology is the capability of a given drug to exhibit activity with respect to multiple drug targets, which are not necessarily in the same activity class. Adverse drug reactions ('side effects') are its principal manifestation, but polypharmacology is also playing a role in the repositioning of existing drugs for new therapeutic indications.  Polyspecificity, on the other hand, is the capability of a given target to exhibit activity with respect to multiple, structurally dissimilar drugs.  That these concepts are closely related to one another is, surprisingly, not well known.  It will be shown in this work that they are, in fact, mathematically related to one another and are in essence 'two sides of the same coin'.  Hence, information on polypharmacology provides equivalent information on polyspecificity, and vice versa. Networks are playing an increasingly important role in biological research. Drug-target networks, in particular, are made up of drug nodes that are linked to specific target nodes if a given drug is active with respect to that target.  Such networks provide a graphic depiction of polypharmacology and polyspecificity.  However, by their very nature they can obscure information that may be useful in their interpretation and analysis.  This work will show how such latent information can be used to determine bounds for the degrees of polypharmacology and polyspecificity, and how to estimate other useful features associated with the lack of completeness of most drug-target datasets.

ChemSAR: an online pipelining platform for molecular SAR modeling

Background

In recent years, predictive models based on machine learning techniques have proven to be feasible and effective in drug discovery. However, to develop such a model, researchers usually have to combine multiple tools and undergo several different steps (e.g., RDKit or ChemoPy package for molecular descriptor calculation, ChemAxon Standardizer for structure preprocessing, scikit-learn package for model building, and ggplot2 package for statistical analysis and visualization, etc.). In addition, it may require strong programming skills to accomplish these jobs, which poses severe challenges for users without advanced training in computer programming. Therefore, an online pipelining platform that integrates a number of selected tools is a valuable and efficient solution that can meet the needs of related researchers.

Results

This work presents a web-based pipelining platform, called ChemSAR, for generating SAR classification models of small molecules. The capabilities of ChemSAR include the validation and standardization of chemical structure representation, the computation of 783 1D/2D molecular descriptors and ten types of widely-used fingerprints for small molecules, the filtering methods for feature selection, the generation of predictive models via a step-by-step job submission process, model interpretation in terms of feature importance and tree visualization, as well as a helpful report generation system. The results can be visualized as high-quality plots and downloaded as local files.

Conclusion

ChemSAR provides an integrated web-based platform for generating SAR classification models that will benefit cheminformatics and other biomedical users. It is freely available at: http://chemsar.scbdd.com.Graphical abstract

The impact of in silico screening in the discovery of novel and safer drug candidates

Drug discovery is a multidisciplinary and multivariate optimization endeavor. As such, in silico screening tools have gained considerable importance to archive, analyze and exploit the vast and ever-increasing amount of experimental data generated throughout the process. The current review will focus on the computer-aided prediction of the numerous properties that need to be controlled during the discovery of a preliminary hit and its promotion to a viable clinical candidate. It does not pretend to the almost impossible task of an exhaustive report but will highlight a few key points that need to be collectively addressed both by chemists and biologists to fuel the drug discovery pipeline with innovative and safe drug candidates.

Comparing Drug Images and Repurposing Drugs with BioGPS and FLAPdock: The Thymidylate Synthase Case

Repurposing and repositioning drugs has become a frequently pursued and successful strategy in the current era, as new chemical entities are increasingly difficult to find and get approved. Herein we report an integrated BioGPS/FLAPdock pipeline for rapid and effective off-target identification and drug repurposing. Our method is based on the structural and chemical properties of protein binding sites, that is, the ligand image, encoded in the GRID molecular interaction fields (MIFs). Protein similarity is disclosed through the BioGPS algorithm by measuring the pockets' overlap according to which pockets are clustered. Co-crystallized and known ligands can be cross-docked among similar targets, selected for subsequent in vitro binding experiments, and possibly improved for inhibitory potency. We used human thymidylate synthase (TS) as a test case and searched the entire RCSB Protein Data Bank (PDB) for similar target pockets. We chose casein kinase IIα as a control and tested a series of its inhibitors against the TS template. Ellagic acid and apigenin were identified as TS inhibitors, and various flavonoids were selected and synthesized in a second-round selection. The compounds were demonstrated to be active in the low-micromolar range.