Link Prediction on a Network of Co-occurring MeSH Terms: Towards Literature-based Discovery

Graph embedding-based link prediction for literature-based discovery in Alzheimer's Disease

OBJECTIVE

We explore the framing of literature-based discovery (LBD) as link prediction and graph embedding learning, with Alzheimer's Disease (AD) as our focus disease context. The key link prediction setting of prediction window length is specifically examined in the context of a time-sliced evaluation methodology.

METHODS

We propose a four-stage approach to explore literature-based discovery for Alzheimer's Disease, creating and analyzing a knowledge graph tailored to the AD context, and predicting and evaluating new knowledge based on time-sliced link prediction. The first stage is to collect an AD-specific corpus. The second stage involves constructing an AD knowledge graph with identified AD-specific concepts and relations from the corpus. In the third stage, 20 pairs of training and testing datasets are constructed with the time-slicing methodology. Finally, we infer new knowledge with graph embedding-based link prediction methods. We compare different link prediction methods in this context. The impact of limiting prediction evaluation of LBD models in the context of short-term and longer-term knowledge evolution for Alzheimer's Disease is assessed.

RESULTS

We constructed an AD corpus of over 16 k papers published in 1977-2021, and automatically annotated it with concepts and relations covering 11 AD-specific semantic entity types. The knowledge graph of Alzheimer's Disease derived from this resource consisted of ∼11 k nodes and ∼394 k edges, among which 34% were genotype-phenotype relationships, 57% were genotype-genotype relationships, and 9% were phenotype-phenotype relationships. A Structural Deep Network Embedding (SDNE) model consistently showed the best performance in terms of returning the most confident set of link predictions as time progresses over 20 years. A huge improvement in model performance was observed when changing the link prediction evaluation setting to consider a more distant future, reflecting the time required for knowledge accumulation.

CONCLUSION

Neural network graph-embedding link prediction methods show promise for the literature-based discovery context, although the prediction setting is extremely challenging, with graph densities of less than 1%. Varying prediction window length on the time-sliced evaluation methodology leads to hugely different results and interpretations of LBD studies. Our approach can be generalized to enable knowledge discovery for other diseases.

AVAILABILITY

Code, AD ontology, and data are available at https://github.com/READ-BioMed/readbiomed-lbd.

Combining Literature Mining and Machine Learning for Predicting Biomedical Discoveries

The major outcomes and insights of scientific research and clinical study end up in the form of publication or clinical record in an unstructured text format. Due to advancements in biomedical research, the growth of published literature is getting tremendous large in recent years. The scientists and clinical researchers are facing a big challenge to stay current with the knowledge and to extract hidden information from this sheer quantity of millions of published biomedical literature. The potential one-stop automated solution to this problem is biomedical literature mining. One of the long-standing goals in biology is to discover the disease-causing genes and their specific roles in personalized precision medicine and drug repurposing. However, the empirical approaches and clinical affirmation are expensive and time-consuming. In silico approach using text mining to identify the disease causing genes can contribute towards biomarker discovery. This chapter presents a protocol on combining literature mining and machine learning for predicting biomedical discoveries with a special emphasis on gene-disease relation based discovery. The protocol is presented as a literature based discovery (LBD) pipeline for gene-disease based discovery. The protocol includes our web based tools: (1) DNER (Disease Named Entity Recognizer) for disease entity recognition, (2) BCCNER (Bidirectional, Contextual clues Named Entity Tagger) for gene/protein entity recognition, (3) DisGeReExT (Disease-Gene Relation Extractor) for statistically validated results and visualization, and (4) a newly introduced deep learning based method for association discovery. Our proposed deep learning based method can be generalized and applied to other important biomedical discoveries focusing on entities such as drug/chemical, or miRNA.

Using Literature Based Discovery to Gain Insights Into the Metabolomic Processes of Cardiac Arrest

In this paper, we describe how we applied LBD techniques to discover lecithin cholesterol acyltransferase (LCAT) as a druggable target for cardiac arrest. We fully describe our process which includes the use of high-throughput metabolomic analysis to identify metabolites significantly related to cardiac arrest, and how we used LBD to gain insights into how these metabolites relate to cardiac arrest. These insights lead to our proposal (for the first time) of LCAT as a druggable target; the effects of which are supported by in vivo studies which were brought forth by this work. Metabolites are the end product of many biochemical pathways within the human body. Observed changes in metabolite levels are indicative of changes in these pathways, and provide valuable insights toward the cause, progression, and treatment of diseases. Following cardiac arrest, we observed changes in metabolite levels pre- and post-resuscitation. We used LBD to help discover diseases implicitly linked via these metabolites of interest. Results of LBD indicated a strong link between Fish Eye disease and cardiac arrest. Since fish eye disease is characterized by an LCAT deficiency, it began an investigation into the effects of LCAT and cardiac arrest survival. In the investigation, we found that decreased LCAT activity may increase cardiac arrest survival rates by increasing ω-3 polyunsaturated fatty acid availability in circulation. We verified the effects of ω-3 polyunsaturated fatty acids on increasing survival rate following cardiac arrest via in vivo with rat models.

Recent advances in biomedical literature mining

The recent years have witnessed a rapid increase in the number of scientific articles in biomedical domain. These literature are mostly available and readily accessible in electronic format. The domain knowledge hidden in them is critical for biomedical research and applications, which makes biomedical literature mining (BLM) techniques highly demanding. Numerous efforts have been made on this topic from both biomedical informatics (BMI) and computer science (CS) communities. The BMI community focuses more on the concrete application problems and thus prefer more interpretable and descriptive methods, while the CS community chases more on superior performance and generalization ability, thus more sophisticated and universal models are developed. The goal of this paper is to provide a review of the recent advances in BLM from both communities and inspire new research directions.

PubMed-Scale Chemical Concept Embeddings Reconstruct Physical Protein Interaction Networks

PubMed is the largest resource of curated biomedical knowledge to date, entailing more than 25 million documents. Large quantities of novel literature prevent a single expert from keeping track of all potentially relevant papers, resulting in knowledge gaps. In this article, we present CHEMMESHNET, a newly developed PubMed-based network comprising more than 10,000,000 associations, constructed from expert-curated MeSH annotations of chemicals based on all currently available PubMed articles. By learning latent representations of concepts in the obtained network, we demonstrate in a proof of concept study that purely literature-based representations are sufficient for the reconstruction of a large part of the currently known network of physical, empirically determined protein-protein interactions. We demonstrate that simple linear embeddings of node pairs, when coupled with a neural network-based classifier, reliably reconstruct the existing collection of empirically confirmed protein-protein interactions. Furthermore, we demonstrate how pairs of learned representations can be used to prioritize potentially interesting novel interactions based on the common chemical context. Highly ranked interactions are qualitatively inspected in terms of potential complex formation at the structural level and represent potentially interesting new knowledge. We demonstrate that two protein-protein interactions, prioritized by structure-based approaches, also emerge as probable with regard to the trained machine-learning model.

Seven-Layer Model in Complex Networks Link Prediction: A Survey

Link prediction is the most basic and essential problem in complex networks. This study analyzes the observed topological, time, attributive, label, weight, directional, and symbolic features and auxiliary information to find the lack of connection and predict the future possible connection. For discussion and analysis of the evolution of the network, the network model is of great significance. In the past two decades, link prediction has attracted extensive attention from experts in various fields, who have published numerous high-level papers, but few combine interdisciplinary characteristics. This survey analyzes and discusses the existing link prediction methods. The idea of stratification is introduced into the classification system of link prediction for the first time and proposes the design idea of a seven-layer model, namely the network, metadata, feature classification, selection input, processing, selection, and output layers. Among them, the processing layer divides link prediction methods into similarity-based, probabilistic, likelihood, supervised learning, semi-supervised learning, unsupervised learning, and reinforcement learning methods. The input features, evaluation metrics, complex analysis, experimental comparisons, relative merits, common dataset and open-source implementations for each link prediction method are then discussed in detail. Through analysis and comparison, we found that the link prediction method based on graph structure features has better prediction performance. Finally, the future development direction of link prediction in complex networks is discussed.

Tracking and Mining the COVID-19 Research Literature

The unprecedented, explosive growth of the COVID-19 domain presents challenges to researchers to keep up with research knowledge within the domain. This article profiles this research to help make that knowledge more accessible via overviews and novel categorizations. We provide websites offering means for researchers to probe more deeply to address specific questions. We further probe and reassemble COVID-19 topical content to address research issues concerning topical evolution and emphases on tactical vs. strategic approaches to mitigate this pandemic and reduce future viral threats. Data suggest that heightened attention to strategic, immunological factors is warranted. Connecting with and transferring in research knowledge from outside the COVID-19 domain demand a viable COVID-19 knowledge model. This study provides complementary topical categorizations to facilitate such modeling to inform future Literature-Based Discovery endeavors.

Inferring new relations between medical entities using literature curated term co-occurrences

Objectives

Identifying new relations between medical entities, such as drugs, diseases, and side effects, is typically a resource-intensive task, involving experimentation and clinical trials. The increased availability of related data and curated knowledge enables a computational approach to this task, notably by training models to predict likely relations. Such models rely on meaningful representations of the medical entities being studied. We propose a generic features vector representation that leverages co-occurrences of medical terms, linked with PubMed citations.

Materials and Methods

We demonstrate the usefulness of the proposed representation by inferring two types of relations: a drug causes a side effect and a drug treats an indication. To predict these relations and assess their effectiveness, we applied 2 modeling approaches: multi-task modeling using neural networks and single-task modeling based on gradient boosting machines and logistic regression.

Results

These trained models, which predict either side effects or indications, obtained significantly better results than baseline models that use a single direct co-occurrence feature. The results demonstrate the advantage of a comprehensive representation.

Discussion

Selecting the appropriate representation has an immense impact on the predictive performance of machine learning models. Our proposed representation is powerful, as it spans multiple medical domains and can be used to predict a wide range of relation types.

Conclusion

The discovery of new relations between various medical entities can be translated into meaningful insights, for example, related to drug development or disease understanding. Our representation of medical entities can be used to train models that predict such relations, thus accelerating healthcare-related discoveries.

Novel Compound-Target Interactions Prediction for the Herbal Formula Hua-Yu-Qiang-Shen-Tong-Bi-Fang

Herbal formulae have a long history in clinical medicine in Asia. While the complexity of the formulae leads to the complex compound-target interactions and the resultant multi-target therapeutic effects, it is difficult to elucidate the molecular/therapeutic mechanism of action for the many formulae. For example, the Hua-Yu-Qiang-Shen-Tong-Bi-Fang (TBF), an herbal formula of Chinese medicine, has been used for treating rheumatoid arthritis. However, the target information of a great number of compounds from the TBF formula is missing. In this study, we predicted the targets of the compounds from the TBF formula via network analysis and in silico computing. Initially, the information of the phytochemicals contained in the plants of the herbal formula was collected, and subsequently computed to their corresponding fingerprints for the sake of structural similarity calculation. Then a compound structural similarity network infused with available target information was constructed. Five local similarity indices were used and compared for their performance on predicting the potential new targets of the compounds. Finally, the Preferential Attachment Index was selected for it having an area under curve (AUC) of 0.886, which outperforms the other four algorithms in predicting the compound-target interactions. This method could provide a promising direction for identifying the compound-target interactions of herbal formulae in silico.

Indirect association and ranking hypotheses for literature based discovery

BACKGROUND

Literature Based Discovery (LBD) produces more potential hypotheses than can be manually reviewed, making automatically ranking these hypotheses critical. In this paper, we introduce the indirect association measures of Linking Term Association (LTA), Minimum Weight Association (MWA), and Shared B to C Set Association (SBC), and compare them to Linking Set Association (LSA), concept embeddings vector cosine, Linking Term Count (LTC), and direct co-occurrence vector cosine. Our proposed indirect association measures extend traditional association measures to quantify indirect rather than direct associations while preserving valuable statistical properties.

RESULTS

We perform a comparison between several different hypothesis ranking methods for LBD, and compare them against our proposed indirect association measures. We intrinsically evaluate each method's performance using its ability to estimate semantic relatedness on standard evaluation datasets. We extrinsically evaluate each method's ability to rank hypotheses in LBD using a time-slicing dataset based on co-occurrence information, and another time-slicing dataset based on SemRep extracted-relationships. Precision and recall curves are generated by ranking term pairs and applying a threshold at each rank.

CONCLUSIONS

Results differ depending on the evaluation methods and datasets, but it is unclear if this is a result of biases in the evaluation datasets or if one method is truly better than another. We conclude that LTC and SBC are the best suited methods for hypothesis ranking in LBD, but there is value in having a variety of methods to choose from.

Towards self-learning based hypotheses generation in biomedical text domain

Motivation

The overwhelming amount of research articles in the domain of bio-medicine might cause important connections to remain unnoticed. Literature Based Discovery is a sub-field within biomedical text mining that peruses these articles to formulate high confident hypotheses on possible connections between medical concepts. Although many alternate methodologies have been proposed over the last decade, they still suffer from scalability issues. The primary reason, apart from the dense inter-connections between biological concepts, is the absence of information on the factors that lead to the edge-formation. In this work, we formulate this problem as a collaborative filtering task and leverage a relatively new concept of word-vectors to learn and mimic the implicit edge-formation process. Along with single-class classifier, we prune the search-space of redundant and irrelevant hypotheses to increase the efficiency of the system and at the same time maintaining and in some cases even boosting the overall accuracy.

Results

We show that our proposed framework is able to prune up to 90% of the hypotheses while still retaining high recall in top-K results. This level of efficiency enables the discovery algorithm to look for higher-order hypotheses, something that was infeasible until now. Furthermore, the generic formulation allows our approach to be agile to perform both open and closed discovery. We also experimentally validate that the core data-structures upon which the system bases its decision has a high concordance with the opinion of the experts.This coupled with the ability to understand the edge formation process provides us with interpretable results without any manual intervention.

Availability and implementation

The relevant JAVA codes are available at: https://github.com/vishrawas/Medline-Code_v2.

Supplementary information

Supplementary data are available at Bioinformatics online.

A survey on literature based discovery approaches in biomedical domain

Literature Based Discovery (LBD) refers to the problem of inferring new and interesting knowledge by logically connecting independent fragments of information units through explicit or implicit means. This area of research, which incorporates techniques from Natural Language Processing (NLP), Information Retrieval and Artificial Intelligence, has significant potential to reduce discovery time in biomedical research fields. Formally introduced in 1986, LBD has grown to be a significant and a core task for text mining practitioners in the biomedical domain. Together with its inter-disciplinary nature, this has led researchers across domains to contribute in advancing this field of study. This survey attempts to consolidate and present the evolution of techniques in this area. We cover a variety of techniques and provide a detailed description of the problem setting, the intuition, the advantages and limitations of various influential papers. We also list the current bottlenecks in this field and provide a general direction of research activities for the future. In an effort to be comprehensive and for ease of reference for off-the-shelf users, we also list many publicly available tools for LBD. We hope this survey will act as a guide to both academic and industry (bio)-informaticians, introduce the various methodologies currently employed and also the challenges yet to be tackled.

A systematic review on literature-based discovery workflow

As scientific publication rates increase, knowledge acquisition and the research development process have become more complex and time-consuming. Literature-Based Discovery (LBD), supporting automated knowledge discovery, helps facilitate this process by eliciting novel knowledge by analysing existing scientific literature. This systematic review provides a comprehensive overview of the LBD workflow by answering nine research questions related to the major components of the LBD workflow (i.e., input, process, output, and evaluation). With regards to the input component, we discuss the data types and data sources used in the literature. The process component presents filtering techniques, ranking/thresholding techniques, domains, generalisability levels, and resources. Subsequently, the output component focuses on the visualisation techniques used in LBD discipline. As for the evaluation component, we outline the evaluation techniques, their generalisability, and the quantitative measures used to validate results. To conclude, we summarise the findings of the review for each component by highlighting the possible future research directions.

Disentangling the evolution of MEDLINE bibliographic database: A complex network perspective

Scientific knowledge constitutes a complex system that has recently been the topic of in-depth analysis. Empirical evidence reveals that little is known about the dynamic aspects of human knowledge. Precise dissection of the expansion of scientific knowledge could help us to better understand the evolutionary dynamics of science. In this paper, we analyzed the dynamic properties and growth principles of the MEDLINE bibliographic database using network analysis methodology. The basic assumption of this work is that the scientific evolution of the life sciences can be represented as a list of co-occurrences of MeSH descriptors that are linked to MEDLINE citations. The MEDLINE database was summarized as a complex system, consisting of nodes and edges, where the nodes refer to knowledge concepts and the edges symbolize corresponding relations. We performed an extensive statistical evaluation based on more than 25 million citations in the MEDLINE database, from 1966 until 2014. We based our analysis on node and community level in order to track temporal evolution in the network. The degree distribution of the network follows a stretched exponential distribution which prevents the creation of large hubs. Results showed that the appearance of new MeSH terms does not also imply new connections. The majority of new connections among nodes results from old MeSH descriptors. We suggest a wiring mechanism based on the theory of structural holes, according to which a novel scientific discovery is established when a connection is built among two or more previously disconnected parts of scientific knowledge. Overall, we extracted 142 different evolving communities. It is evident that new communities are constantly born, live for some time, and then die. We also provide a Web-based application that helps characterize and understand the content of extracted communities. This study clearly shows that the evolution of MEDLINE knowledge correlates with the network's structural and temporal characteristics.

Creation of Individual Scientific Concept-Centered Semantic Maps Based on Automated Text-Mining Analysis of PubMed

Concept-centered semantic maps were created based on a text-mining analysis of PubMed using the BiblioEngine_v2018 software. The objects ("concepts") of a semantic map can be MeSH-terms or other terms (names of proteins, diseases, chemical compounds, etc.) structured in the form of controlled vocabularies. The edges between the two objects were automatically calculated based on the index of semantic similarity, which is proportional to the number of publications related to both objects simultaneously. On the one hand, an individual semantic map created based on the already published papers allows us to trace scientific inquiry. On the other hand, a prospective analysis based on the study of PubMed search history enables us to determine the possible directions for future research.

Neural networks for link prediction in realistic biomedical graphs: a multi-dimensional evaluation of graph embedding-based approaches

Background

Link prediction in biomedical graphs has several important applications including predicting Drug-Target Interactions (DTI), Protein-Protein Interaction (PPI) prediction and Literature-Based Discovery (LBD). It can be done using a classifier to output the probability of link formation between nodes. Recently several works have used neural networks to create node representations which allow rich inputs to neural classifiers. Preliminary works were done on this and report promising results. However they did not use realistic settings like time-slicing, evaluate performances with comprehensive metrics or explain when or why neural network methods outperform. We investigated how inputs from four node representation algorithms affect performance of a neural link predictor on random- and time-sliced biomedical graphs of real-world sizes (∼ 6 million edges) containing information relevant to DTI, PPI and LBD. We compared the performance of the neural link predictor to those of established baselines and report performance across five metrics.

Results

In random- and time-sliced experiments when the neural network methods were able to learn good node representations and there was a negligible amount of disconnected nodes, those approaches outperformed the baselines. In the smallest graph (∼ 15,000 edges) and in larger graphs with approximately 14% disconnected nodes, baselines such as Common Neighbours proved a justifiable choice for link prediction. At low recall levels (∼ 0.3) the approaches were mostly equal, but at higher recall levels across all nodes and average performance at individual nodes, neural network approaches were superior. Analysis showed that neural network methods performed well on links between nodes with no previous common neighbours; potentially the most interesting links. Additionally, while neural network methods benefit from large amounts of data, they require considerable amounts of computational resources to utilise them.

Conclusions

Our results indicate that when there is enough data for the neural network methods to use and there are a negligible amount of disconnected nodes, those approaches outperform the baselines. At low recall levels the approaches are mostly equal but at higher recall levels and average performance at individual nodes, neural network approaches are superior. Performance at nodes without common neighbours which indicate more unexpected and perhaps more useful links account for this.

Electronic supplementary material

The online version of this article (10.1186/s12859-018-2163-9) contains supplementary material, which is available to authorized users.

Predicting potential drug-drug interactions on topological and semantic similarity features using statistical learning

Drug-drug interaction (DDI) is a change in the effect of a drug when patient takes another drug. Characterizing DDIs is extremely important to avoid potential adverse drug reactions. We represent DDIs as a complex network in which nodes refer to drugs and links refer to their potential interactions. Recently, the problem of link prediction has attracted much consideration in scientific community. We represent the process of link prediction as a binary classification task on networks of potential DDIs. We use link prediction techniques for predicting unknown interactions between drugs in five arbitrary chosen large-scale DDI databases, namely DrugBank, KEGG, NDF-RT, SemMedDB, and Twosides. We estimated the performance of link prediction using a series of experiments on DDI networks. We performed link prediction using unsupervised and supervised approach including classification tree, k-nearest neighbors, support vector machine, random forest, and gradient boosting machine classifiers based on topological and semantic similarity features. Supervised approach clearly outperforms unsupervised approach. The Twosides network gained the best prediction performance regarding the area under the precision-recall curve (0.93 for both random forests and gradient boosting machine). The applied methodology can be used as a tool to help researchers to identify potential DDIs. The supervised link prediction approach proved to be promising for potential DDIs prediction and may facilitate the identification of potential DDIs in clinical research.

Rediscovering Don Swanson: the Past, Present and Future of Literature-Based Discovery

The late Don R. Swanson was well appreciated during his lifetime as Dean of the Graduate Library School at University of Chicago, as winner of the American Society for Information Science Award of Merit for 2000, and as author of many seminal articles. In this informal essay, I will give my personal perspective on Don's contributions to science, and outline some current and future directions in literature-based discovery that are rooted in concepts that he developed.

Gaps within the Biomedical Literature: Initial Characterization and Assessment of Strategies for Discovery

Within well-established fields of biomedical science, we identify "gaps", topical areas of investigation that might be expected to occur but are missing. We define a field by carrying out a topical PubMed query, and analyze Medical Subject Headings by which the set of retrieved articles are indexed. Medical Subject headings (MeSH terms) which occur in >1% of the articles are examined pairwise to see how often they are predicted to co-occur within individual articles (assuming that they are independent of each other). A pair of MeSH terms that are predicted to co-occur in at least 10 articles, yet are not observed to co-occur in any article, are "gaps" and were studied further in a corpus of 10 disease-related article sets and 10 related to biological processes. Overall, articles that filled gaps were cited more heavily than non-gap-filling articles and were 61% more likely to be published in multidisciplinary high-impact journals. Nine different features of these "gaps" were characterized and tested to learn which, if any, correlate with the appearance of one or more articles containing both MeSH terms within the next five years. Several different types of gaps were identified, each having distinct combinations of predictive features: a) those arising as a byproduct of MeSH indexing rules; b) those having little biological meaning; c) those representing "low hanging fruit" for immediate exploitation; and d) those representing gaps across disciplines or sub-disciplines that do not talk to each other or work together. We have built a free, open tool called "Mine the Gap!" that identifies and characterizes the "gaps" for any PubMed query, which can be accessed via the Anne O'Tate value-added PubMed search interface (http://arrowsmith.psych.uic.edu/cgi-bin/arrowsmith_uic/AnneOTate.cgi).