Combining biomedical knowledge graphs and text to improve predictions for drug-target interactions and drug-indications

Knowledge Graphs in Pharmacovigilance: A Scoping Review

PURPOSE

To critically assess the role and added value of knowledge graphs in pharmacovigilance, focusing on their ability to predict adverse drug reactions.

METHODS

A systematic scoping review was conducted in which detailed information, including objectives, technology, data sources, methodology, and performance metrics, were extracted from a set of peer-reviewed publications reporting the use of knowledge graphs to support pharmacovigilance signal detection.

FINDINGS

The review, which included 47 peer-reviewed articles, found knowledge graphs were utilized for detecting/predicting single-drug adverse reactions and drug-drug interactions, with variable reported performance and sparse comparisons to legacy methods.

IMPLICATIONS

Research to date suggests that knowledge graphs have the potential to augment predictive signal detection in pharmacovigilance, but further research using more reliable reference sets of adverse drug reactions and comparison with legacy pharmacovigilance methods are needed to more clearly define best practices and to establish their place in holistic pharmacovigilance systems.

BioBLP: a modular framework for learning on multimodal biomedical knowledge graphs

BACKGROUND

Knowledge graphs (KGs) are an important tool for representing complex relationships between entities in the biomedical domain. Several methods have been proposed for learning embeddings that can be used to predict new links in such graphs. Some methods ignore valuable attribute data associated with entities in biomedical KGs, such as protein sequences, or molecular graphs. Other works incorporate such data, but assume that entities can be represented with the same data modality. This is not always the case for biomedical KGs, where entities exhibit heterogeneous modalities that are central to their representation in the subject domain.

OBJECTIVE

We aim to understand how to incorporate multimodal data into biomedical KG embeddings, and analyze the resulting performance in comparison with traditional methods. We propose a modular framework for learning embeddings in KGs with entity attributes, that allows encoding attribute data of different modalities while also supporting entities with missing attributes. We additionally propose an efficient pretraining strategy for reducing the required training runtime. We train models using a biomedical KG containing approximately 2 million triples, and evaluate the performance of the resulting entity embeddings on the tasks of link prediction, and drug-protein interaction prediction, comparing against methods that do not take attribute data into account.

RESULTS

In the standard link prediction evaluation, the proposed method results in competitive, yet lower performance than baselines that do not use attribute data. When evaluated in the task of drug-protein interaction prediction, the method compares favorably with the baselines. Further analyses show that incorporating attribute data does outperform baselines over entities below a certain node degree, comprising approximately 75% of the diseases in the graph. We also observe that optimizing attribute encoders is a challenging task that increases optimization costs. Our proposed pretraining strategy yields significantly higher performance while reducing the required training runtime.

CONCLUSION

BioBLP allows to investigate different ways of incorporating multimodal biomedical data for learning representations in KGs. With a particular implementation, we find that incorporating attribute data does not consistently outperform baselines, but improvements are obtained on a comparatively large subset of entities below a specific node-degree. Our results indicate a potential for improved performance in scientific discovery tasks where understudied areas of the KG would benefit from link prediction methods.

Toxicology knowledge graph for structural birth defects

BACKGROUND

Birth defects are functional and structural abnormalities that impact about 1 in 33 births in the United States. They have been attributed to genetic and other factors such as drugs, cosmetics, food, and environmental pollutants during pregnancy, but for most birth defects there are no known causes.

METHODS

To further characterize associations between small molecule compounds and their potential to induce specific birth abnormalities, we gathered knowledge from multiple sources to construct a reproductive toxicity Knowledge Graph (ReproTox-KG) with a focus on associations between birth defects, drugs, and genes. Specifically, we gathered data from drug/birth-defect associations from co-mentions in published abstracts, gene/birth-defect associations from genetic studies, drug- and preclinical-compound-induced gene expression changes in cell lines, known drug targets, genetic burden scores for human genes, and placental crossing scores for small molecules.

RESULTS

Using ReproTox-KG and semi-supervised learning (SSL), we scored >30,000 preclinical small molecules for their potential to cross the placenta and induce birth defects, and identified >500 birth-defect/gene/drug cliques that can be used to explain molecular mechanisms for drug-induced birth defects. The ReproTox-KG can be accessed via a web-based user interface available at https://maayanlab.cloud/reprotox-kg . This site enables users to explore the associations between birth defects, approved and preclinical drugs, and all human genes.

CONCLUSIONS

ReproTox-KG provides a resource for exploring knowledge about the molecular mechanisms of birth defects with the potential of predicting the likelihood of genes and preclinical small molecules to induce birth defects.

OncoRTT: Predicting novel oncology-related therapeutic targets using BERT embeddings and omics features

Late-stage drug development failures are usually a consequence of ineffective targets. Thus, proper target identification is needed, which may be possible using computational approaches. The reason being, effective targets have disease-relevant biological functions, and omics data unveil the proteins involved in these functions. Also, properties that favor the existence of binding between drug and target are deducible from the protein’s amino acid sequence. In this work, we developed OncoRTT, a deep learning (DL)-based method for predicting novel therapeutic targets. OncoRTT is designed to reduce suboptimal target selection by identifying novel targets based on features of known effective targets using DL approaches. First, we created the “OncologyTT” datasets, which include genes/proteins associated with ten prevalent cancer types. Then, we generated three sets of features for all genes: omics features, the proteins’ amino-acid sequence BERT embeddings, and the integrated features to train and test the DL classifiers separately. The models achieved high prediction performances in terms of area under the curve (AUC), i.e., AUC greater than 0.88 for all cancer types, with a maximum of 0.95 for leukemia. Also, OncoRTT outperformed the state-of-the-art method using their data in five out of seven cancer types commonly assessed by both methods. Furthermore, OncoRTT predicts novel therapeutic targets using new test data related to the seven cancer types. We further corroborated these results with other validation evidence using the Open Targets Platform and a case study focused on the top-10 predicted therapeutic targets for lung cancer.

Knowledge-Driven Mechanistic Enrichment of the Preeclampsia Ignorome

Preeclampsia is a leading cause of maternal and fetal morbidity and mortality. Currently, the only definitive treatment of preeclampsia is delivery of the placenta, which is central to the pathogenesis of the disease. Transcriptional profiling of human placenta from pregnancies complicated by preeclampsia has been extensively performed to identify differentially expressed genes (DEGs). The decisions to investigate DEGs experimentally are biased by many factors, causing many DEGs to remain uninvestigated. A set of DEGs which are associated with a disease experimentally, but which have no known association to the disease in the literature are known as the ignorome. Preeclampsia has an extensive body of scientific literature, a large pool of DEG data, and only one definitive treatment. Tools facilitating knowledge-based analyses, which are capable of combining disparate data from many sources in order to suggest underlying mechanisms of action, may be a valuable resource to support discovery and improve our understanding of this disease. In this work we demonstrate how a biomedical knowledge graph (KG) can be used to identify novel preeclampsia molecular mechanisms. Existing open source biomedical resources and publicly available high-throughput transcriptional profiling data were used to identify and annotate the function of currently uninvestigated preeclampsia-associated DEGs. Experimentally investigated genes associated with preeclampsia were identified from PubMed abstracts using text-mining methodologies. The relative complement of the text-mined- and meta-analysis-derived lists were identified as the uninvestigated preeclampsia-associated DEGs (n=445), i.e., the preeclampsia ignorome. Using the KG to investigate relevant DEGs revealed 53 novel clinically relevant and biologically actionable mechanistic associations.

Prediction of Drug–Target Interaction Using Dual-Network Integrated Logistic Matrix Factorization and Knowledge Graph Embedding

Nowadays, drug–target interactions (DTIs) prediction is a fundamental part of drug repositioning. However, on the one hand, drug–target interactions prediction models usually consider drugs or targets information, which ignore prior knowledge between drugs and targets. On the other hand, models incorporating priori knowledge cannot make interactions prediction for under-studied drugs and targets. Hence, this article proposes a novel dual-network integrated logistic matrix factorization DTIs prediction scheme (Ro-DNILMF) via a knowledge graph embedding approach. This model adds prior knowledge as input data into the prediction model and inherits the advantages of the DNILMF model, which can predict under-studied drug–target interactions. Firstly, a knowledge graph embedding model based on relational rotation (RotatE) is trained to construct the interaction adjacency matrix and integrate prior knowledge. Secondly, a dual-network integrated logistic matrix factorization prediction model (DNILMF) is used to predict new drugs and targets. Finally, several experiments conducted on the public datasets are used to demonstrate that the proposed method outperforms the single base-line model and some mainstream methods on efficiency.