MScanner: a classifier for retrieving Medline citations

A comparative evaluation of biomedical similar article recommendation

BACKGROUND

Biomedical sciences, with their focus on human health and disease, have attracted unprecedented attention in the 21st century. The proliferation of biomedical sciences has also led to a large number of scientific articles being produced, which makes it difficult for biomedical researchers to find relevant articles and hinders the dissemination of valuable discoveries. To bridge this gap, the research community has initiated the article recommendation task, with the aim of recommending articles to biomedical researchers automatically based on their research interests. Over the past two decades, many recommendation methods have been developed. However, an algorithm-level comparison and rigorous evaluation of the most important methods on a shared dataset is still lacking.

METHOD

In this study, we first investigate 15 methods for automated article recommendation in the biomedical domain. We then conduct an empirical evaluation of the 15 methods, including six term-based methods, two word embedding methods, three sentence embedding methods, two document embedding methods, and two BERT-based methods. These methods are evaluated in two scenarios: article-oriented recommenders and user-oriented recommenders, with two publicly available datasets: TREC 2005 Genomics and RELISH, respectively.

RESULTS

Our experimental results show that the text representation models BERT and BioSenVec outperform many existing recommendation methods (e.g., BM25, PMRA, XPRC) and web-based recommendation systems (e.g., MScanner, MedlineRanker, BioReader) on both datasets regarding most of the evaluation metrics, and fine-tuning can improve the performance of the BERT-based methods.

CONCLUSIONS

Our comparison study is useful for researchers and practitioners in selecting the best modeling strategies for building article recommendation systems in the biomedical domain. The code and datasets are publicly available.

BioReader: a text mining tool for performing classification of biomedical literature

Background

Scientific data and research results are being published at an unprecedented rate. Many database curators and researchers utilize data and information from the primary literature to populate databases, form hypotheses, or as the basis for analyses or validation of results. These efforts largely rely on manual literature surveys for collection of these data, and while querying the vast amounts of literature using keywords is enabled by repositories such as PubMed, filtering relevant articles from such query results can be a non-trivial and highly time consuming task.

Results

We here present a tool that enables users to perform classification of scientific literature by text mining-based classification of article abstracts. BioReader (Biomedical Research Article Distiller) is trained by uploading article corpora for two training categories - e.g. one positive and one negative for content of interest - as well as one corpus of abstracts to be classified and/or a search string to query PubMed for articles. The corpora are submitted as lists of PubMed IDs and the abstracts are automatically downloaded from PubMed, preprocessed, and the unclassified corpus is classified using the best performing classification algorithm out of ten implemented algorithms.

Conclusion

BioReader supports data and information collection by implementing text mining-based classification of primary biomedical literature in a web interface, thus enabling curators and researchers to take advantage of the vast amounts of data and information in the published literature. BioReader outperforms existing tools with similar functionalities and expands the features used for mining literature in database curation efforts. The tool is freely available as a web service at http://www.cbs.dtu.dk/services/BioReader

Electronic supplementary material

The online version of this article (10.1186/s12859-019-2607-x) contains supplementary material, which is available to authorized users.

Large expert-curated database for benchmarking document similarity detection in biomedical literature search

Document recommendation systems for locating relevant literature have mostly relied on methods developed a decade ago. This is largely due to the lack of a large offline gold-standard benchmark of relevant documents that cover a variety of research fields such that newly developed literature search techniques can be compared, improved and translated into practice. To overcome this bottleneck, we have established the RElevant LIterature SearcH consortium consisting of more than 1500 scientists from 84 countries, who have collectively annotated the relevance of over 180 000 PubMed-listed articles with regard to their respective seed (input) article/s. The majority of annotations were contributed by highly experienced, original authors of the seed articles. The collected data cover 76% of all unique PubMed Medical Subject Headings descriptors. No systematic biases were observed across different experience levels, research fields or time spent on annotations. More importantly, annotations of the same document pairs contributed by different scientists were highly concordant. We further show that the three representative baseline methods used to generate recommended articles for evaluation (Okapi Best Matching 25, Term Frequency-Inverse Document Frequency and PubMed Related Articles) had similar overall performances. Additionally, we found that these methods each tend to produce distinct collections of recommended articles, suggesting that a hybrid method may be required to completely capture all relevant articles. The established database server located at https://relishdb.ict.griffith.edu.au is freely available for the downloading of annotation data and the blind testing of new methods. We expect that this benchmark will be useful for stimulating the development of new powerful techniques for title and title/abstract-based search engines for relevant articles in biomedical research.

Large expert-curated database for benchmarking document similarity detection in biomedical literature search

Document recommendation systems for locating relevant literature have mostly relied on methods developed a decade ago. This is largely due to the lack of a large offline gold-standard benchmark of relevant documents that cover a variety of research fields such that newly developed literature search techniques can be compared, improved and translated into practice. To overcome this bottleneck, we have established the RElevant LIterature SearcH consortium consisting of more than 1500 scientists from 84 countries, who have collectively annotated the relevance of over 180 000 PubMed-listed articles with regard to their respective seed (input) article/s. The majority of annotations were contributed by highly experienced, original authors of the seed articles. The collected data cover 76% of all unique PubMed Medical Subject Headings descriptors. No systematic biases were observed across different experience levels, research fields or time spent on annotations. More importantly, annotations of the same document pairs contributed by different scientists were highly concordant. We further show that the three representative baseline methods used to generate recommended articles for evaluation (Okapi Best Matching 25, Term Frequency-Inverse Document Frequency and PubMed Related Articles) had similar overall performances. Additionally, we found that these methods each tend to produce distinct collections of recommended articles, suggesting that a hybrid method may be required to completely capture all relevant articles. The established database server located at https://relishdb.ict.griffith.edu.au is freely available for the downloading of annotation data and the blind testing of new methods. We expect that this benchmark will be useful for stimulating the development of new powerful techniques for title and title/abstract-based search engines for relevant articles in biomedical research.

Information Retrieval and Text Mining Technologies for Chemistry

Efficient access to chemical information contained in scientific literature, patents, technical reports, or the web is a pressing need shared by researchers and patent attorneys from different chemical disciplines. Retrieval of important chemical information in most cases starts with finding relevant documents for a particular chemical compound or family. Targeted retrieval of chemical documents is closely connected to the automatic recognition of chemical entities in the text, which commonly involves the extraction of the entire list of chemicals mentioned in a document, including any associated information. In this Review, we provide a comprehensive and in-depth description of fundamental concepts, technical implementations, and current technologies for meeting these information demands. A strong focus is placed on community challenges addressing systems performance, more particularly CHEMDNER and CHEMDNER patents tasks of BioCreative IV and V, respectively. Considering the growing interest in the construction of automatically annotated chemical knowledge bases that integrate chemical information and biological data, cheminformatics approaches for mapping the extracted chemical names into chemical structures and their subsequent annotation together with text mining applications for linking chemistry with biological information are also presented. Finally, future trends and current challenges are highlighted as a roadmap proposal for research in this emerging field.

Mining biomedical images towards valuable information retrieval in biomedical and life sciences

Biomedical images are helpful sources for the scientists and practitioners in drawing significant hypotheses, exemplifying approaches and describing experimental results in published biomedical literature. In last decades, there has been an enormous increase in the amount of heterogeneous biomedical image production and publication, which results in a need for bioimaging platforms for feature extraction and analysis of text and content in biomedical images to take advantage in implementing effective information retrieval systems. In this review, we summarize technologies related to data mining of figures. We describe and compare the potential of different approaches in terms of their developmental aspects, used methodologies, produced results, achieved accuracies and limitations. Our comparative conclusions include current challenges for bioimaging software with selective image mining, embedded text extraction and processing of complex natural language queries.

Finding Related Publications: Extending the Set of Terms Used to Assess Article Similarity

Recommendation of related articles is an important feature of the PubMed. The PubMed Related Citations (PRC) algorithm is the engine that enables this feature, and it leverages information on 22 million citations. We analyzed the performance of the PRC algorithm on 4584 annotated articles from the 2005 Text REtrieval Conference (TREC) Genomics Track data. Our analysis indicated that the PRC highest weighted term was not always consistent with the critical term that was most directly related to the topic of the article. We implemented term expansion and found that it was a promising and easy-to-implement approach to improve the performance of the PRC algorithm for the TREC 2005 Genomics data and for the TREC 2014 Clinical Decision Support Track data. For term expansion, we trained a Skip-gram model using the Word2Vec package. This extended PRC algorithm resulted in higher average precision for a large subset of articles. A combination of both algorithms may lead to improved performance in related article recommendations.

Integrating unified medical language system and association mining techniques into relevance feedback for biomedical literature search

Background

Finding highly relevant articles from biomedical databases is challenging not only because it is often difficult to accurately express a user’s underlying intention through keywords but also because a keyword-based query normally returns a long list of hits with many citations being unwanted by the user. This paper proposes a novel biomedical literature search system, called BiomedSearch, which supports complex queries and relevance feedback.

Methods

The system employed association mining techniques to build a k-profile representing a user’s relevance feedback. More specifically, we developed a weighted interest measure and an association mining algorithm to find the strength of association between a query and each concept in the article(s) selected by the user as feedback. The top concepts were utilized to form a k-profile used for the next-round search. BiomedSearch relies on Unified Medical Language System (UMLS) knowledge sources to map text files to standard biomedical concepts. It was designed to support queries with any levels of complexity.

Results

A prototype of BiomedSearch software was made and it was preliminarily evaluated using the Genomics data from TREC (Text Retrieval Conference) 2006 Genomics Track. Initial experiment results indicated that BiomedSearch increased the mean average precision (MAP) for a set of queries.

Conclusions

With UMLS and association mining techniques, BiomedSearch can effectively utilize users’ relevance feedback to improve the performance of biomedical literature search.

MSL: Facilitating automatic and physical analysis of published scientific literature in PDF format

Published scientific literature contains millions of figures, including information about the results obtained from different scientific experiments e.g. PCR-ELISA data, microarray analysis, gel electrophoresis, mass spectrometry data, DNA/RNA sequencing, diagnostic imaging (CT/MRI and ultrasound scans), and medicinal imaging like electroencephalography (EEG), magnetoencephalography (MEG), echocardiography  (ECG), positron-emission tomography (PET) images. The importance of biomedical figures has been widely recognized in scientific and medicine communities, as they play a vital role in providing major original data, experimental and computational results in concise form. One major challenge for implementing a system for scientific literature analysis is extracting and analyzing text and figures from published PDF files by physical and logical document analysis. Here we present a product line architecture based bioinformatics tool ‘Mining Scientific Literature (MSL)’, which supports the extraction of text and images by interpreting all kinds of published PDF files using advanced data mining and image processing techniques. It provides modules for the marginalization of extracted text based on different coordinates and keywords, visualization of extracted figures and extraction of embedded text from all kinds of biological and biomedical figures using applied Optimal Character Recognition (OCR). Moreover, for further analysis and usage, it generates the system’s output in different formats including text, PDF, XML and images files. Hence, MSL is an easy to install and use analysis tool to interpret published scientific literature in PDF format.

Text mining for neuroanatomy using WhiteText with an updated corpus and a new web application

We describe the WhiteText project, and its progress towards automatically extracting statements of neuroanatomical connectivity from text. We review progress to date on the three main steps of the project: recognition of brain region mentions, standardization of brain region mentions to neuroanatomical nomenclature, and connectivity statement extraction. We further describe a new version of our manually curated corpus that adds 2,111 connectivity statements from 1,828 additional abstracts. Cross-validation classification within the new corpus replicates results on our original corpus, recalling 67% of connectivity statements at 51% precision. The resulting merged corpus provides 5,208 connectivity statements that can be used to seed species-specific connectivity matrices and to better train automated techniques. Finally, we present a new web application that allows fast interactive browsing of the over 70,000 sentences indexed by the system, as a tool for accessing the data and assisting in further curation. Software and data are freely available at http://www.chibi.ubc.ca/WhiteText/.

Comparing the Precision of Information Retrieval of MeSH-Controlled Vocabulary Search Method and a Visual Method in the Medline Medical Database

BACKGROUND

Medline is one of the most important databases in the biomedical field. One of the most important hosts for Medline is Elton B. Stephens CO. (EBSCO), which has presented different search methods that can be used based on the needs of the users. Visual search and MeSH-controlled search methods are among the most common methods. The goal of this research was to compare the precision of the retrieved sources in the EBSCO Medline base using MeSH-controlled and visual search methods.

METHODS

This research was a semi-empirical study. By holding training workshops, 70 students of higher education in different educational departments of Kashan University of Medical Sciences were taught MeSH-Controlled and visual search methods in 2012. Then, the precision of 300 searches made by these students was calculated based on Best Precision, Useful Precision, and Objective Precision formulas and analyzed in SPSS software using the independent sample T Test, and three precisions obtained with the three precision formulas were studied for the two search methods.

RESULTS

The mean precision of the visual method was greater than that of the MeSH-Controlled search for all three types of precision, i.e. Best Precision, Useful Precision, and Objective Precision, and their mean precisions were significantly different (P <0.001). Sixty-five percent of the researchers indicated that, although the visual method was better than the controlled method, the control of keywords in the controlled method resulted in finding more proper keywords for the searches. Fifty-three percent of the participants in the research also mentioned that the use of the combination of the two methods produced better results.

CONCLUSION

For users, it is more appropriate to use a natural, language-based method, such as the visual method, in the EBSCO Medline host than to use the controlled method, which requires users to use special keywords. The potential reason for their preference was that the visual method allowed them more freedom of action.

BioTextQuest(+): a knowledge integration platform for literature mining and concept discovery

SUMMARY

The iterative process of finding relevant information in biomedical literature and performing bioinformatics analyses might result in an endless loop for an inexperienced user, considering the exponential growth of scientific corpora and the plethora of tools designed to mine PubMed(®) and related biological databases. Herein, we describe BioTextQuest(+), a web-based interactive knowledge exploration platform with significant advances to its predecessor (BioTextQuest), aiming to bridge processes such as bioentity recognition, functional annotation, document clustering and data integration towards literature mining and concept discovery. BioTextQuest(+) enables PubMed and OMIM querying, retrieval of abstracts related to a targeted request and optimal detection of genes, proteins, molecular functions, pathways and biological processes within the retrieved documents. The front-end interface facilitates the browsing of document clustering per subject, the analysis of term co-occurrence, the generation of tag clouds containing highly represented terms per cluster and at-a-glance popup windows with information about relevant genes and proteins. Moreover, to support experimental research, BioTextQuest(+) addresses integration of its primary functionality with biological repositories and software tools able to deliver further bioinformatics services. The Google-like interface extends beyond simple use by offering a range of advanced parameterization for expert users. We demonstrate the functionality of BioTextQuest(+) through several exemplary research scenarios including author disambiguation, functional term enrichment, knowledge acquisition and concept discovery linking major human diseases, such as obesity and ageing.

AVAILABILITY

The service is accessible at http://bioinformatics.med.uoc.gr/biotextquest.

CONTACT

g.pavlopoulos@gmail.com or georgios.pavlopoulos@esat.kuleuven.be

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

How to learn about gene function: text-mining or ontologies?

As the amount of genome information increases rapidly, there is a correspondingly greater need for methods that provide accurate and automated annotation of gene function. For example, many high-throughput technologies--e.g., next-generation sequencing--are being used today to generate lists of genes associated with specific conditions. However, their functional interpretation remains a challenge and many tools exist trying to characterize the function of gene-lists. Such systems rely typically in enrichment analysis and aim to give a quick insight into the underlying biology by presenting it in a form of a summary-report. While the load of annotation may be alleviated by such computational approaches, the main challenge in modern annotation remains to develop a systems form of analysis in which a pipeline can effectively analyze gene-lists quickly and identify aggregated annotations through computerized resources. In this article we survey some of the many such tools and methods that have been developed to automatically interpret the biological functions underlying gene-lists. We overview current functional annotation aspects from the perspective of their epistemology (i.e., the underlying theories used to organize information about gene function into a body of verified and documented knowledge) and find that most of the currently used functional annotation methods fall broadly into one of two categories: they are based either on 'known' formally-structured ontology annotations created by 'experts' (e.g., the GO terms used to describe the function of Entrez Gene entries), or--perhaps more adventurously--on annotations inferred from literature (e.g., many text-mining methods use computer-aided reasoning to acquire knowledge represented in natural languages). Overall however, deriving detailed and accurate insight from such gene lists remains a challenging task, and improved methods are called for. In particular, future methods need to (1) provide more holistic insight into the underlying molecular systems; (2) provide better follow-up experimental testing and treatment options, and (3) better manage gene lists derived from organisms that are not well-studied. We discuss some promising approaches that may help achieve these advances, especially the use of extended dictionaries of biomedical concepts and molecular mechanisms, as well as greater use of annotation benchmarks.

Area under precision-recall curves for weighted and unweighted data

Precision-recall curves are highly informative about the performance of binary classifiers, and the area under these curves is a popular scalar performance measure for comparing different classifiers. However, for many applications class labels are not provided with absolute certainty, but with some degree of confidence, often reflected by weights or soft labels assigned to data points. Computing the area under the precision-recall curve requires interpolating between adjacent supporting points, but previous interpolation schemes are not directly applicable to weighted data. Hence, even in cases where weights were available, they had to be neglected for assessing classifiers using precision-recall curves. Here, we propose an interpolation for precision-recall curves that can also be used for weighted data, and we derive conditions for classification scores yielding the maximum and minimum area under the precision-recall curve. We investigate accordances and differences of the proposed interpolation and previous ones, and we demonstrate that taking into account existing weights of test data is important for the comparison of classifiers.

Accessing biomedical literature in the current information landscape

Biomedical and life sciences literature is unique because of its exponentially increasing volume and interdisciplinary nature. Biomedical literature access is essential for several types of users including biomedical researchers, clinicians, database curators, and bibliometricians. In the past few decades, several online search tools and literature archives, generic as well as biomedicine specific, have been developed. We present this chapter in the light of three consecutive steps of literature access: searching for citations, retrieving full text, and viewing the article. The first section presents the current state of practice of biomedical literature access, including an analysis of the search tools most frequently used by the users, including PubMed, Google Scholar, Web of Science, Scopus, and Embase, and a study on biomedical literature archives such as PubMed Central. The next section describes current research and the state-of-the-art systems motivated by the challenges a user faces during query formulation and interpretation of search results. The research solutions are classified into five key areas related to text and data mining, text similarity search, semantic search, query support, relevance ranking, and clustering results. Finally, the last section describes some predicted future trends for improving biomedical literature access, such as searching and reading articles on portable devices, and adoption of the open access policy.

Biological information extraction and co-occurrence analysis

Nowadays, it is possible to identify terms corresponding to biological entities within passages in biomedical text corpora: critically, their potential relationships then need to be detected. These relationships are typically detected by co-occurrence analysis, revealing associations between bioentities through their coexistence in single sentences and/or entire abstracts. These associations implicitly define networks, whose nodes represent terms/bioentities/concepts being connected by relationship edges; edge weights might represent confidence for these semantic connections.This chapter provides a review of current methods for co-occurrence analysis, focusing on data storage, analysis, and representation. We highlight scenarios of these approaches implemented by useful tools for information extraction and knowledge inference in the field of systems biology. We illustrate the practical utility of two online resources providing services of this type-namely, STRING and BioTextQuest-concluding with a discussion of current challenges and future perspectives in the field.

Comparison and combination of several MeSH indexing approaches

MeSH indexing of MEDLINE is becoming a more difficult task for the group of highly qualified indexing staff at the US National Library of Medicine, due to the large yearly growth of MEDLINE and the increasing size of MeSH. Since 2002, this task has been assisted by the Medical Text Indexer or MTI program. We extend previous machine learning analysis by adding a more diverse set of MeSH headings targeting examples where MTI has been shown to perform poorly. Machine learning algorithms exceed MTI's performance on MeSH headings that are used very frequently and headings for which the indexing frequency is very low. We find that when we combine the MTI suggestions and the prediction of the learning algorithms, the performance improves compared to any single method for most of the evaluated MeSH headings.

Text Categorization of Heart, Lung, and Blood Studies in the Database of Genotypes and Phenotypes (dbGaP) Utilizing n-grams and Metadata Features

The database of Genotypes and Phenotypes (dbGaP) allows researchers to understand phenotypic contribution to genetic conditions, generate new hypotheses, confirm previous study results, and identify control populations. However, effective use of the database is hindered by suboptimal study retrieval. Our objective is to evaluate text classification techniques to improve study retrieval in the context of the dbGaP database. We utilized standard machine learning algorithms (naive Bayes, support vector machines, and the C4.5 decision tree) trained on dbGaP study text and incorporated n-gram features and study metadata to identify heart, lung, and blood studies. We used the χ(2) feature selection algorithm to identify features that contributed most to classification performance and experimented with dbGaP associated PubMed papers as a proxy for topicality. Classifier performance was favorable in comparison to keyword-based search results. It was determined that text categorization is a useful complement to document retrieval techniques in the dbGaP.

MeSH indexing based on automatically generated summaries

BACKGROUND

MEDLINE citations are manually indexed at the U.S. National Library of Medicine (NLM) using as reference the Medical Subject Headings (MeSH) controlled vocabulary. For this task, the human indexers read the full text of the article. Due to the growth of MEDLINE, the NLM Indexing Initiative explores indexing methodologies that can support the task of the indexers. Medical Text Indexer (MTI) is a tool developed by the NLM Indexing Initiative to provide MeSH indexing recommendations to indexers. Currently, the input to MTI is MEDLINE citations, title and abstract only. Previous work has shown that using full text as input to MTI increases recall, but decreases precision sharply. We propose using summaries generated automatically from the full text for the input to MTI to use in the task of suggesting MeSH headings to indexers. Summaries distill the most salient information from the full text, which might increase the coverage of automatic indexing approaches based on MEDLINE. We hypothesize that if the results were good enough, manual indexers could possibly use automatic summaries instead of the full texts, along with the recommendations of MTI, to speed up the process while maintaining high quality of indexing results.

RESULTS

We have generated summaries of different lengths using two different summarizers, and evaluated the MTI indexing on the summaries using different algorithms: MTI, individual MTI components, and machine learning. The results are compared to those of full text articles and MEDLINE citations. Our results show that automatically generated summaries achieve similar recall but higher precision compared to full text articles. Compared to MEDLINE citations, summaries achieve higher recall but lower precision.

CONCLUSIONS

Our results show that automatic summaries produce better indexing than full text articles. Summaries produce similar recall to full text but much better precision, which seems to indicate that automatic summaries can efficiently capture the most important contents within the original articles. The combination of MEDLINE citations and automatically generated summaries could improve the recommendations suggested by MTI. On the other hand, indexing performance might be dependent on the MeSH heading being indexed. Summarization techniques could thus be considered as a feature selection algorithm that might have to be tuned individually for each MeSH heading.

Using cited references to improve the retrieval of related biomedical documents

BACKGROUND

A popular query from scientists reading a biomedical abstract is to search for topic-related documents in bibliographic databases. Such a query is challenging because the amount of information attached to a single abstract is little, whereas classification-based retrieval algorithms are optimally trained with large sets of relevant documents. As a solution to this problem, we propose a query expansion method that extends the information related to a manuscript using its cited references.

RESULTS

Data on cited references and text sections in 249,108 full-text biomedical articles was extracted from the Open Access subset of the PubMed Central® database (PMC-OA). Of the five standard sections of a scientific article, the Introduction and Discussion sections contained most of the citations (mean = 10.2 and 9.9 citations, respectively). A large proportion of articles (98.4%) and their cited references (79.5%) were indexed in the PubMed® database. Using the MedlineRanker abstract classification tool, cited references allowed accurate retrieval of the citing document in a test set of 10,000 documents and also of documents related to six biomedical topics defined by particular MeSH® terms from the entire PMC-OA (p-value<0.01). Classification performance was sensitive to the topic and also to the text sections from which the references were selected. Classifiers trained on the baseline (i.e., only text from the query document and not from the references) were outperformed in almost all the cases. Best performance was often obtained when using all cited references, though using the references from Introduction and Discussion sections led to similarly good results. This query expansion method performed significantly better than pseudo relevance feedback in 4 out of 6 topics.

CONCLUSIONS

The retrieval of documents related to a single document can be significantly improved by using the references cited by this document (p-value<0.01). Using references from Introduction and Discussion performs almost as well as using all references, which might be useful for methods that require reduced datasets due to computational limitations. Cited references from particular sections might not be appropriate for all topics. Our method could be a better alternative to pseudo relevance feedback though it is limited by full text availability.

Application and evaluation of automated methods to extract neuroanatomical connectivity statements from free text

MOTIVATION

Automated annotation of neuroanatomical connectivity statements from the neuroscience literature would enable accessible and large-scale connectivity resources. Unfortunately, the connectivity findings are not formally encoded and occur as natural language text. This hinders aggregation, indexing, searching and integration of the reports. We annotated a set of 1377 abstracts for connectivity relations to facilitate automated extraction of connectivity relationships from neuroscience literature. We tested several baseline measures based on co-occurrence and lexical rules. We compare results from seven machine learning methods adapted from the protein interaction extraction domain that employ part-of-speech, dependency and syntax features.

RESULTS

Co-occurrence based methods provided high recall with weak precision. The shallow linguistic kernel recalled 70.1% of the sentence-level connectivity statements at 50.3% precision. Owing to its speed and simplicity, we applied the shallow linguistic kernel to a large set of new abstracts. To evaluate the results, we compared 2688 extracted connections with the Brain Architecture Management System (an existing database of rat connectivity). The extracted connections were connected in the Brain Architecture Management System at a rate of 63.5%, compared with 51.1% for co-occurring brain region pairs. We found that precision increases with the recency and frequency of the extracted relationships.

AVAILABILITY AND IMPLEMENTATION

The source code, evaluations, documentation and other supplementary materials are available at http://www.chibi.ubc.ca/WhiteText.

CONTACT

paul@chibi.ubc.ca.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics Online.

Caipirini: using gene sets to rank literature

BACKGROUND

Keeping up-to-date with bioscience literature is becoming increasingly challenging. Several recent methods help meet this challenge by allowing literature search to be launched based on lists of abstracts that the user judges to be 'interesting'. Some methods go further by allowing the user to provide a second input set of 'uninteresting' abstracts; these two input sets are then used to search and rank literature by relevance. In this work we present the service 'Caipirini' (http://caipirini.org) that also allows two input sets, but takes the novel approach of allowing ranking of literature based on one or more sets of genes.

RESULTS

To evaluate the usefulness of Caipirini, we used two test cases, one related to the human cell cycle, and a second related to disease defense mechanisms in Arabidopsis thaliana. In both cases, the new method achieved high precision in finding literature related to the biological mechanisms underlying the input data sets.

CONCLUSIONS

To our knowledge Caipirini is the first service enabling literature search directly based on biological relevance to gene sets; thus, Caipirini gives the research community a new way to unlock hidden knowledge from gene sets derived via high-throughput experiments.

Cost sensitive hierarchical document classification to triage PubMed abstracts for manual curation

BACKGROUND

The Immune Epitope Database (IEDB) project manually curates information from published journal articles that describe immune epitopes derived from a wide variety of organisms and associated with different diseases. In the past, abstracts of scientific articles were retrieved by broad keyword queries of PubMed, and were classified as relevant (curatable) or irrelevant (not curatable) to the scope of the database by a Naïve Bayes classifier. The curatable abstracts were subsequently manually classified into categories corresponding to different disease domains. Over the past four years, we have examined how to further improve this approach in order to enhance classification performance and to reduce the need for manual intervention.

RESULTS

Utilizing 89,884 abstracts classified by a domain expert as curatable or uncuratable, we found that a SVM classifier outperformed the previously used Naïve Bayes classifier for curatability predictions with an AUC of 0.899 and 0.854, respectively. Next, using a non-hierarchical and a hierarchical application of SVM classifiers trained on 22,833 curatable abstracts manually classified into three levels of disease specific categories we demonstrated that a hierarchical application of SVM classifiers outperformed non-hierarchical SVM classifiers for categorization. Finally, to optimize the hierarchical SVM classifiers' error profile for the curation process, cost sensitivity functions were developed to avoid serious misclassifications. We tested our design on a benchmark dataset of 1,388 references and achieved an overall category prediction accuracy of 94.4%, 93.9%, and 82.1% at the three levels of categorization, respectively.

CONCLUSIONS

A hierarchical application of SVM algorithms with cost sensitive output weighting enabled high quality reference classification with few serious misclassifications. This enabled us to significantly reduce the manual component of abstract categorization. Our findings are relevant to other databases that are developing their own document classifier schema and the datasets we make available provide large scale real-life benchmark sets for method developers.

A bottom-up approach to MEDLINE indexing recommendations

MEDLINE indexing performed by the US National Library of Medicine staff describes the essence of a biomedical publication in about 14 Medical Subject Headings (MeSH). Since 2002, this task is assisted by the Medical Text Indexer (MTI) program. We present a bottom-up approach to MEDLINE indexing in which the abstract is searched for indicators for a specific MeSH recommendation in a two-step process. Supervised machine learning combined with triage rules improves sensitivity of recommendations while keeping the number of recommended terms relatively small. Improvement in recommendations observed in this work warrants further exploration of this approach to MTI recommendations on a larger set of MeSH headings.

The Protein-Protein Interaction tasks of BioCreative III: classification/ranking of articles and linking bio-ontology concepts to full text

BACKGROUND

Determining usefulness of biomedical text mining systems requires realistic task definition and data selection criteria without artificial constraints, measuring performance aspects that go beyond traditional metrics. The BioCreative III Protein-Protein Interaction (PPI) tasks were motivated by such considerations, trying to address aspects including how the end user would oversee the generated output, for instance by providing ranked results, textual evidence for human interpretation or measuring time savings by using automated systems. Detecting articles describing complex biological events like PPIs was addressed in the Article Classification Task (ACT), where participants were asked to implement tools for detecting PPI-describing abstracts. Therefore the BCIII-ACT corpus was provided, which includes a training, development and test set of over 12,000 PPI relevant and non-relevant PubMed abstracts labeled manually by domain experts and recording also the human classification times. The Interaction Method Task (IMT) went beyond abstracts and required mining for associations between more than 3,500 full text articles and interaction detection method ontology concepts that had been applied to detect the PPIs reported in them.

RESULTS

A total of 11 teams participated in at least one of the two PPI tasks (10 in ACT and 8 in the IMT) and a total of 62 persons were involved either as participants or in preparing data sets/evaluating these tasks. Per task, each team was allowed to submit five runs offline and another five online via the BioCreative Meta-Server. From the 52 runs submitted for the ACT, the highest Matthew's Correlation Coefficient (MCC) score measured was 0.55 at an accuracy of 89% and the best AUC iP/R was 68%. Most ACT teams explored machine learning methods, some of them also used lexical resources like MeSH terms, PSI-MI concepts or particular lists of verbs and nouns, some integrated NER approaches. For the IMT, a total of 42 runs were evaluated by comparing systems against manually generated annotations done by curators from the BioGRID and MINT databases. The highest AUC iP/R achieved by any run was 53%, the best MCC score 0.55. In case of competitive systems with an acceptable recall (above 35%) the macro-averaged precision ranged between 50% and 80%, with a maximum F-Score of 55%.

CONCLUSIONS

The results of the ACT task of BioCreative III indicate that classification of large unbalanced article collections reflecting the real class imbalance is still challenging. Nevertheless, text-mining tools that report ranked lists of relevant articles for manual selection can potentially reduce the time needed to identify half of the relevant articles to less than 1/4 of the time when compared to unranked results. Detecting associations between full text articles and interaction detection method PSI-MI terms (IMT) is more difficult than might be anticipated. This is due to the variability of method term mentions, errors resulting from pre-processing of articles provided as PDF files, and the heterogeneity and different granularity of method term concepts encountered in the ontology. However, combining the sophisticated techniques developed by the participants with supporting evidence strings derived from the articles for human interpretation could result in practical modules for biological annotation workflows.

Recent progress in automatically extracting information from the pharmacogenomic literature

The biomedical literature holds our understanding of pharmacogenomics, but it is dispersed across many journals. In order to integrate our knowledge, connect important facts across publications and generate new hypotheses we must organize and encode the contents of the literature. By creating databases of structured pharmocogenomic knowledge, we can make the value of the literature much greater than the sum of the individual reports. We can, for example, generate candidate gene lists or interpret surprising hits in genome-wide association studies. Text mining automatically adds structure to the unstructured knowledge embedded in millions of publications, and recent years have seen a surge in work on biomedical text mining, some specific to pharmacogenomics literature. These methods enable extraction of specific types of information and can also provide answers to general, systemic queries. In this article, we describe the main tasks of text mining in the context of pharmacogenomics, summarize recent applications and anticipate the next phase of text mining applications.

MeSHy: Mining unanticipated PubMed information using frequencies of occurrences and concurrences of MeSH terms

MOTIVATION

PubMed is the most widely used database of biomedical literature. To the detriment of the user though, the ranking of the documents retrieved for a query is not content-based, and important semantic information in the form of assigned Medical Subject Headings (MeSH) terms is not readily presented or productively utilized. The motivation behind this work was the discovery of unanticipated information through the appropriate ranking of MeSH term pairs and, indirectly, documents. Such information can be useful in guiding novel research and following promising trends.

METHODS

A web-based tool, called MeSHy, was developed implementing a mainly statistical algorithm. The algorithm takes into account the frequencies of occurrences, concurrences, and the semantic similarities of MeSH terms in retrieved PubMed documents to create MeSH term pairs. These are then scored and ranked, focusing on their unexpectedly frequent or infrequent occurrences.

RESULTS

MeSHy presents results through an online interactive interface facilitating further manipulation through filtering and sorting. The results themselves include the MeSH term pairs, along with MeSH categories, the score, and document IDs, all of which are hyperlinked for convenience. To highlight the applicability of the tool, we report the findings of an expert in the pharmacology field on querying the molecularly-targeted drug imatinib and nutrition-related flavonoids. To the best of our knowledge, MeSHy is the first publicly available tool able to directly provide such a different perspective on the complex nature of published work.

IMPLEMENTATION AND AVAILABILITY

Implemented in Perl and served by Apache2 at http://bat.ina.certh.gr/tools/meshy/ with all major browsers supported.

Clustering more than two million biomedical publications: comparing the accuracies of nine text-based similarity approaches

Background

We investigate the accuracy of different similarity approaches for clustering over two million biomedical documents. Clustering large sets of text documents is important for a variety of information needs and applications such as collection management and navigation, summary and analysis. The few comparisons of clustering results from different similarity approaches have focused on small literature sets and have given conflicting results. Our study was designed to seek a robust answer to the question of which similarity approach would generate the most coherent clusters of a biomedical literature set of over two million documents.

Methodology

We used a corpus of 2.15 million recent (2004-2008) records from MEDLINE, and generated nine different document-document similarity matrices from information extracted from their bibliographic records, including titles, abstracts and subject headings. The nine approaches were comprised of five different analytical techniques with two data sources. The five analytical techniques are cosine similarity using term frequency-inverse document frequency vectors (tf-idf cosine), latent semantic analysis (LSA), topic modeling, and two Poisson-based language models – BM25 and PMRA (PubMed Related Articles). The two data sources were a) MeSH subject headings, and b) words from titles and abstracts. Each similarity matrix was filtered to keep the top-n highest similarities per document and then clustered using a combination of graph layout and average-link clustering. Cluster results from the nine similarity approaches were compared using (1) within-cluster textual coherence based on the Jensen-Shannon divergence, and (2) two concentration measures based on grant-to-article linkages indexed in MEDLINE.

Conclusions

PubMed's own related article approach (PMRA) generated the most coherent and most concentrated cluster solution of the nine text-based similarity approaches tested, followed closely by the BM25 approach using titles and abstracts. Approaches using only MeSH subject headings were not competitive with those based on titles and abstracts.

PubMed and beyond: a survey of web tools for searching biomedical literature

The past decade has witnessed the modern advances of high-throughput technology and rapid growth of research capacity in producing large-scale biological data, both of which were concomitant with an exponential growth of biomedical literature. This wealth of scholarly knowledge is of significant importance for researchers in making scientific discoveries and healthcare professionals in managing health-related matters. However, the acquisition of such information is becoming increasingly difficult due to its large volume and rapid growth. In response, the National Center for Biotechnology Information (NCBI) is continuously making changes to its PubMed Web service for improvement. Meanwhile, different entities have devoted themselves to developing Web tools for helping users quickly and efficiently search and retrieve relevant publications. These practices, together with maturity in the field of text mining, have led to an increase in the number and quality of various Web tools that provide comparable literature search service to PubMed. In this study, we review 28 such tools, highlight their respective innovations, compare them to the PubMed system and one another, and discuss directions for future development. Furthermore, we have built a website dedicated to tracking existing systems and future advances in the field of biomedical literature search. Taken together, our work serves information seekers in choosing tools for their needs and service providers and developers in keeping current in the field.

Database URL:http://www.ncbi.nlm.nih.gov/CBBresearch/Lu/search

Enabling multi-level relevance feedback on PubMed by integrating rank learning into DBMS

Background

Finding relevant articles from PubMed is challenging because it is hard to express the user's specific intention in the given query interface, and a keyword query typically retrieves a large number of results. Researchers have applied machine learning techniques to find relevant articles by ranking the articles according to the learned relevance function. However, the process of learning and ranking is usually done offline without integrated with the keyword queries, and the users have to provide a large amount of training documents to get a reasonable learning accuracy. This paper proposes a novel multi-level relevance feedback system for PubMed, called RefMed, which supports both ad-hoc keyword queries and a multi-level relevance feedback in real time on PubMed.

Results

RefMed supports a multi-level relevance feedback by using the RankSVM as the learning method, and thus it achieves higher accuracy with less feedback. RefMed "tightly" integrates the RankSVM into RDBMS to support both keyword queries and the multi-level relevance feedback in real time; the tight coupling of the RankSVM and DBMS substantially improves the processing time. An efficient parameter selection method for the RankSVM is also proposed, which tunes the RankSVM parameter without performing validation. Thereby, RefMed achieves a high learning accuracy in real time without performing a validation process. RefMed is accessible at http://dm.postech.ac.kr/refmed.

Conclusions

RefMed is the first multi-level relevance feedback system for PubMed, which achieves a high accuracy with less feedback. It effectively learns an accurate relevance function from the user’s feedback and efficiently processes the function to return relevant articles in real time.

Improving the prediction of pharmacogenes using text-derived drug-gene relationships

A critical goal of pharmacogenomics research is to identify genes that can explain variation in drug response. We have previously reported a method that creates a genome-scale ranking of genes likely to interact with a drug. The algorithm uses information about drug structure and indications of use to rank the genes. Although the algorithm has good performance, its performance depends on a curated set of drug-gene relationships that is expensive to create and difficult to maintain. In this work, we assess the utility of text mining in extracting a network of drug-gene relationships automatically. This provides a valuable aggregate source of knowledge, subsequently used as input into the algorithm that ranks potential pharmacogenes. Using a drug-gene network created from sentence-level co-occurrence in the full text of scientific articles, we compared the performance to that of a network created by manual curation of those articles. Under a wide range of conditions, we show that a knowledge base derived from text-mining the literature performs as well as, and sometimes better than, a high-quality, manually curated knowledge base. We conclude that we can use relationships mined automatically from the literature as a knowledgebase for pharmacogenomics relationships. Additionally, when relationships are missed by text mining, our system can accurately extrapolate new relationships with 77.4% precision.

PharmGKB: understanding the effects of individual genetic variants

The Pharmacogenetics and Pharmacogenomics Knowledge Base (PharmGKB: http://www.pharmgkb.org) is devoted to disseminating primary data and knowledge in pharmacogenetics and pharmacogenomics. We are annotating the genes that are most important for drug response and present this information in the form of Very Important Pharmacogene (VIP) summaries, pathway diagrams, and curated literature. The PharmGKB currently contains information on over 500 drugs, 500 diseases, and 700 genes with genotyped variants. New features focus on capturing the phenotypic consequences of individual genetic variants. These features link variant genotypes to phenotypes, increase the breadth of pharmacogenomics literature curated, and visualize single-nucleotide polymorphisms on a gene's three-dimensional protein structure.

MedlineRanker: flexible ranking of biomedical literature

The biomedical literature is represented by millions of abstracts available in the Medline database. These abstracts can be queried with the PubMed interface, which provides a keyword-based Boolean search engine. This approach shows limitations in the retrieval of abstracts related to very specific topics, as it is difficult for a non-expert user to find all of the most relevant keywords related to a biomedical topic. Additionally, when searching for more general topics, the same approach may return hundreds of unranked references. To address these issues, text mining tools have been developed to help scientists focus on relevant abstracts. We have implemented the MedlineRanker webserver, which allows a flexible ranking of Medline for a topic of interest without expert knowledge. Given some abstracts related to a topic, the program deduces automatically the most discriminative words in comparison to a random selection. These words are used to score other abstracts, including those from not yet annotated recent publications, which can be then ranked by relevance. We show that our tool can be highly accurate and that it is able to process millions of abstracts in a practical amount of time. MedlineRanker is free for use and is available at http://cbdm.mdc-berlin.de/tools/medlineranker.

Creating reference datasets for systems biology applications using text mining

High-throughput experimental techniques are generating large data collections with the aim of identifying novel entities involved in fundamental cellular processes as well as drawing a systematic picture of the relationships between individual components. Determining the accuracy of the resulting data and the selection of a subset of targets for more careful characterizations often requires relying on information provided by manually annotated data repositories. These repositories are incomplete and cover only a small fraction of the knowledge contained in the literature. We propose in this paper the use of text-mining technologies to extract, organize, and present information relevant for a particular biological topic. The aims of the resulting approach are (1) to enable topic-centric biological literature navigation, (2) to assist in the construction of manually revised data repositories, (3) to provide prioritization of biological entities for experimental studies, and (4) to enable human interpretation of large-scale experiments by providing direct links of bio-entities to relevant descriptions in the literature.

Linking genes to literature: text mining, information extraction, and retrieval applications for biology

Efficient access to information contained in online scientific literature collections is essential for life science research, playing a crucial role from the initial stage of experiment planning to the final interpretation and communication of the results. The biological literature also constitutes the main information source for manual literature curation used by expert-curated databases. Following the increasing popularity of web-based applications for analyzing biological data, new text-mining and information extraction strategies are being implemented. These systems exploit existing regularities in natural language to extract biologically relevant information from electronic texts automatically. The aim of the BioCreative challenge is to promote the development of such tools and to provide insight into their performance. This review presents a general introduction to the main characteristics and applications of currently available text-mining systems for life sciences in terms of the following: the type of biological information demands being addressed; the level of information granularity of both user queries and results; and the features and methods commonly exploited by these applications. The current trend in biomedical text mining points toward an increasing diversification in terms of application types and techniques, together with integration of domain-specific resources such as ontologies. Additional descriptions of some of the systems discussed here are available on the internet .