Text Mining Pipeline Research Articles

Bioregulatory event extraction using large language models: a case study of rice literature

The extraction of biological regulation events has been a key focus in the field of biomedical nature language processing (BioNLP). However, existing methods often encounter challenges such as cascading errors in text mining pipelines and limitations in topic coverage from the selected corpus. Fortunately, the emergence of large language models (LLMs) presents a potential solution due to their robust semantic understanding and extensive knowledge base. To explore this potential, our project at the Biomedical Linked Annotation Hackathon 8 (BLAH 8) investigates the feasibility of using LLMs to extract biological regulation events. Our findings, based on the analysis of rice literature, demonstrate the promising performance of LLMs in this task, while also highlighting several concerns that must be addressed in future LLM-based application in low-resource topic.

Searching for LINCS to Stress: Using Text Mining to Automate Reference Chemical Curation.

Adaptive stress response pathways (SRPs) restore cellular homeostasis following perturbation but may activate terminal outcomes like apoptosis, autophagy, or cellular senescence if disruption exceeds critical thresholds. Because SRPs hold the key to vital cellular tipping points, they are targeted for therapeutic interventions and assessed as biomarkers of toxicity. Hence, we are developing a public database of chemicals that perturb SRPs to enable new data-driven tools to improve public health. Here, we report on the automated text-mining pipeline we used to build and curate the first version of this database. We started with 100 reference SRP chemicals gathered from published biomarker studies to bootstrap the database. Second, we used information retrieval to find co-occurrences of reference chemicals with SRP terms in PubMed abstracts and determined pairwise mutual information thresholds to filter biologically relevant relationships. Third, we applied these thresholds to find 1206 putative SRP perturbagens within thousands of substances in the Library of Integrated Network-Based Cellular Signatures (LINCS). To assign SRP activity to LINCS chemicals, domain experts had to manually review at least three publications for each of 1206 chemicals out of 181,805 total abstracts. To accomplish this efficiently, we implemented a machine learning approach to predict SRP classifications from texts to prioritize abstracts. In 5-fold cross-validation testing with a corpus derived from the 100 reference chemicals, artificial neural networks performed the best (F1-macro = 0.678) and prioritized 2479/181,805 abstracts for expert review, which resulted in 457 chemicals annotated with SRP activities. An independent analysis of enriched mechanisms of action and chemical use class supported the text-mined chemical associations (p < 0.05): heat shock inducers were linked with HSP90 and DNA damage inducers to topoisomerase inhibition. This database will enable novel applications of LINCS data to evaluate SRP activities and to further develop tools for biomedical information extraction from the literature.

Comorbidity-Guided Text Mining and Omics Pipeline to Identify Candidate Genes and Drugs for Alzheimer's Disease.

Alzheimer's disease (AD), a multifactorial neurodegenerative disorder, is prevalent among the elderly population. It is a complex trait with mutations in multiple genes. Although the US Food and Drug Administration (FDA) has approved a few drugs for AD treatment, a definitive cure remains elusive. Research efforts persist in seeking improved treatment options for AD. Here, a hybrid pipeline is proposed to apply text mining to identify comorbid diseases for AD and an omics approach to identify the common genes between AD and five comorbid diseases-dementia, type 2 diabetes, hypertension, Parkinson's disease, and Down syndrome. We further identified the pathways and drugs for common genes. The rationale behind this approach is rooted in the fact that elderly individuals often receive multiple medications for various comorbid diseases, and an insight into the genes that are common to comorbid diseases may enhance treatment strategies. We identified seven common genes-PSEN1, PSEN2, MAPT, APP, APOE, NOTCH, and HFE-for AD and five comorbid diseases. We investigated the drugs interacting with these common genes using LINCS gene-drug perturbation. Our analysis unveiled several promising candidates, including MG-132 and Masitinib, which exhibit potential efficacy for both AD and its comorbid diseases. The pipeline can be extended to other diseases.

Cancer-Alterome: a literature-mined resource for regulatory events caused by genetic alterations in cancer

It is vital to investigate the complex mechanisms underlying tumors to better understand cancer and develop effective treatments. Metabolic abnormalities and clinical phenotypes can serve as essential biomarkers for diagnosing this challenging disease. Additionally, genetic alterations provide profound insights into the fundamental aspects of cancer. This study introduces Cancer-Alterome, a literature-mined dataset that focuses on the regulatory events of an organism’s biological processes or clinical phenotypes caused by genetic alterations. By proposing and leveraging a text-mining pipeline, we identify 16,681 thousand of regulatory events records encompassing 21K genes, 157K genetic alterations and 154K downstream bio-concepts, extracted from 4,354K pan-cancer literature. The resulting dataset empowers a multifaceted investigation of cancer pathology, enabling the meticulous tracking of relevant literature support. Its potential applications extend to evidence-based medicine and precision medicine, yielding valuable insights for further advancements in cancer research.

Advancing Italian biomedical information extraction with transformers-based models: Methodological insights and multicenter practical application

The introduction of computerized medical records in hospitals has reduced burdensome activities like manual writing and information fetching. However, the data contained in medical records are still far underutilized, primarily because extracting data from unstructured textual medical records takes time and effort. Information Extraction, a subfield of Natural Language Processing, can help clinical practitioners overcome this limitation by using automated text-mining pipelines. In this work, we created the first Italian neuropsychiatric Named Entity Recognition dataset, PsyNIT, and used it to develop a Transformers-based model. Moreover, we collected and leveraged three external independent datasets to implement an effective multicenter model, with overall F1-score 84.77 %, Precision 83.16 %, Recall 86.44 %. The lessons learned are: (i) the crucial role of a consistent annotation process and (ii) a fine-tuning strategy that combines classical methods with a “low-resource” approach. This allowed us to establish methodological guidelines that pave the way for Natural Language Processing studies in less-resourced languages.

DEBBIE: The Open Access Database of Experimental Scaffolds and Biomaterials Built Using an Automated Text Mining Pipeline.

Biomaterials research output has experienced an exponential increase over the last three decades. The majority of research is published in the form of scientific articles and is therefore available as unstructured text, making it a challenging input for computational processing. Computational tools are becoming essential to overcome this information overload. Among them, text mining systems present an attractive option for the automated extraction of information from text documents into structured datasets. This work presents the first automated system for biomaterial related information extraction from the National Library of Medicine's premier bibliographic database (MEDLINE) research abstracts into a searchable database. The system is a text mining pipeline that periodically retrieves abstracts from PubMed and identifies research and clinical studies of biomaterials. Thereafter, the pipeline identifies sixteen concept types of interest in the abstract using the Biomaterials Annotator, a tool for biomaterials Named Entity Recognition (NER). These concepts of interest, along with the abstract and relevant metadata are then deposited in DEBBIE, the Database of Experimental Biomaterials and their Biological Effect. DEBBIE is accessible through a web application that provides keyword searches and displays results in an intuitive and meaningful manner, aiming to facilitate an efficient mapping and organization of biomaterials information.

DETEXA: declarative extensible text exploration and analysis through SQL

Metadata enrichment through text mining techniques is becoming one of the most significant tasks in digital libraries. Due to the exponential increase of open access publications, several new challenges have emerged. Raw data are usually big, unstructured, and come from heterogeneous data sources. In this paper, we introduce a text analysis framework implemented in extended SQL that exploits the scalability characteristics of modern database management systems. The purpose of this framework is to provide the opportunity to build performant end-to-end text mining pipelines which include data harvesting, cleaning, processing, and text analysis at once. SQL is selected due to its declarative nature which offers fast experimentation and the ability to build APIs so that domain experts can edit text mining workflows via easy-to-use graphical interfaces. Our experimental analysis demonstrates that the proposed framework is very effective and achieves significant speedup, up to three times faster, in common use cases compared to other popular approaches.

Neuroimaging-ITM: A Text Mining Pipeline Combining Deep Adversarial Learning with Interaction Based Topic Modeling for Enabling the FAIR Neuroimaging Study.

Sharing various neuroimaging digital resources have received widespread attention in FAIR (Findable, Accessible, Interoperable and Reusable) neuroscience. In order to support a comprehensive understanding of brain cognition, neuroimaging provenance should be constructed to characterize both research processes and results, and integrates various digital resources for quick replication and open cooperation. This brings new challenges to neuroimaging text mining, including fragmented information, lack of labelled corpora, and vague topics. This paper proposes a text mining pipeline for enabling the FAIR neuroimaging study. In order to avoid fragmented information, the Brain Informatics provenance model is redesigned based on NIDM (Neuroimaging Data Model) and FAIR facets. It can systematically capture the provenance requests from the FAIR neuroimaging study and then transform them into a group of text mining tasks. A neuroimaging text mining pipeline combining deep adversarial learning with interaction based topic modeling, called neuroimaging interaction topic model (Neuroimaging-ITM), is proposed to automatically extract neuroimaging provenance and identify research topics in the few-shot scenario. Finally, a group of experiments is completed by using real data from the journal PloS One. The experimental results show that Neuroimaging-ITM can systematically and accurately extract provenance information and obtain high-quality research topics from the full text of neuroimaging articles. Most of the mean F1 values of provenance extraction exceed 0.9. The topic coherence and KL (Kullback-Leibler) divergence reach 9.95 and 0.96 respectively. The results are obviously better than baseline methods.

Automated extraction of genes associated with antibiotic resistance from the biomedical literature.

The detection of bacterial antibiotic resistance phenotypes is important when carrying out clinical decisions for patient treatment. Conventional phenotypic testing involves culturing bacteria which requires a significant amount of time and work. Whole-genome sequencing is emerging as a fast alternative to resistance prediction, by considering the presence/absence of certain genes. A lot of research has focused on determining which bacterial genes cause antibiotic resistance and efforts are being made to consolidate these facts in knowledge bases (KBs). KBs are usually manually curated by domain experts to be of the highest quality. However, this limits the pace at which new facts are added. Automated relation extraction of gene-antibiotic resistance relations from the biomedical literature is one solution that can simplify the curation process. This paper reports on the development of a text mining pipeline that takes in English biomedical abstracts and outputs genes that are predicted to cause resistance to antibiotics. To test the generalisability of this pipeline it was then applied to predict genes associated with Helicobacter pylori antibiotic resistance, that are not present in common antibiotic resistance KBs or publications studying H. pylori. These genes would be candidates for further lab-based antibiotic research and inclusion in these KBs. For relation extraction, state-of-the-art deep learning models were used. These models were trained on a newly developed silver corpus which was generated by distant supervision of abstracts using the facts obtained from KBs. The top performing model was superior to a co-occurrence model, achieving a recall of 95%, a precision of 60% and F1-score of 74% on a manually annotated holdout dataset. To our knowledge, this project was the first attempt at developing a complete text mining pipeline that incorporates deep learning models to extract gene-antibiotic resistance relations from the literature. Additional related data can be found at https://github.com/AndreBrincat/Gene-Antibiotic-Resistance-Relation-Extraction

ExTRI: Extraction of transcription regulation interactions from literature

The regulation of gene transcription by transcription factors is a fundamental biological process, yet the relations between transcription factors (TF) and their target genes (TG) are still only sparsely covered in databases. Text-mining tools can offer broad and complementary solutions to help locate and extract mentions of these biological relationships in articles. We have generated ExTRI, a knowledge graph of TF-TG relationships, by applying a high recall text-mining pipeline to MedLine abstracts identifying over 100,000 candidate sentences with TF-TG relations. Validation procedures indicated that about half of the candidate sentences contain true TF-TG relationships. Post-processing identified 53,000 high confidence sentences containing TF-TG relationships, with a cross-validation F1-score close to 75%. The resulting collection of TF-TG relationships covers 80% of the relations annotated in existing databases. It adds 11,000 other potential interactions, including relationships for ~100 TFs currently not in public TF-TG relation databases. The high confidence abstract sentences contribute 25,000 literature references not available from other resources and offer a wealth of direct pointers to functional aspects of the TF-TG interactions. Our compiled resource encompassing ExTRI together with publicly available resources delivers literature-derived TF-TG interactions for more than 900 of the 1500–1600 proteins considered to function as specific DNA binding TFs. The obtained result can be used by curators, for network analysis and modelling, for causal reasoning or knowledge graph mining approaches, or serve to benchmark text mining strategies.

Contextualizing Genes by Using Text-Mined Co-Occurrence Features for Cancer Gene Panel Discovery.

Developing a biomedical-explainable and validatable text mining pipeline can help in cancer gene panel discovery. We create a pipeline that can contextualize genes by using text-mined co-occurrence features. We apply Biomedical Natural Language Processing (BioNLP) techniques for literature mining in the cancer gene panel. A literature-derived 4,679 × 4,630 gene term-feature matrix was built. The EGFR L858R and T790M, and BRAF V600E genetic variants are important mutation term features in text mining and are frequently mutated in cancer. We validate the cancer gene panel by the mutational landscape of different cancer types. The cosine similarity of gene frequency between text mining and a statistical result from clinical sequencing data is 80.8%. In different machine learning models, the best accuracy for the prediction of two different gene panels, including MSK-IMPACT (Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets), and Oncomine cancer gene panel, is 0.959, and 0.989, respectively. The receiver operating characteristic (ROC) curve analysis confirmed that the neural net model has a better prediction performance (Area under the ROC curve (AUC) = 0.992). The use of text-mined co-occurrence features can contextualize each gene. We believe the approach is to evaluate several existing gene panels, and show that we can use part of the gene panel set to predict the remaining genes for cancer discovery.

Text mining-based word representations for biomedical data analysis and protein-protein interaction networks in machine learning tasks.

Biomedical and life science literature is an essential way to publish experimental results. With the rapid growth of the number of new publications, the amount of scientific knowledge represented in free text is increasing remarkably. There has been much interest in developing techniques that can extract this knowledge and make it accessible to aid scientists in discovering new relationships between biological entities and answering biological questions. Making use of the word2vec approach, we generated word vector representations based on a corpus consisting of over 16 million PubMed abstracts. We developed a text mining pipeline to produce word2vec embeddings with different properties and performed validation experiments to assess their utility for biomedical analysis. An important pre-processing step consisted in the substitution of synonymous terms by their preferred terms in biomedical databases. Furthermore, we extracted gene-gene networks from two embedding versions and used them as prior knowledge to train Graph-Convolutional Neural Networks (CNNs) on large breast cancer gene expression data and on other cancer datasets. Performances of resulting models were compared to Graph-CNNs trained with protein-protein interaction (PPI) networks or with networks derived using other word embedding algorithms. We also assessed the effect of corpus size on the variability of word representations. Finally, we created a web service with a graphical and a RESTful interface to extract and explore relations between biomedical terms using annotated embeddings. Comparisons to biological databases showed that relations between entities such as known PPIs, signaling pathways and cellular functions, or narrower disease ontology groups correlated with higher cosine similarity. Graph-CNNs trained with word2vec-embedding-derived networks performed sufficiently good for the metastatic event prediction tasks compared to other networks. Such performance was good enough to validate the utility of our generated word embeddings in constructing biological networks. Word representations as produced by text mining algorithms like word2vec, therefore are able to capture biologically meaningful relations between entities. Our generated embeddings are publicly available at https://github.com/genexplain/Word2vec-based-Networks/blob/main/README.md.

COVID-19 recommender system based on an annotated multilingual corpus.

Tracking the most recent advances in Coronavirus disease 2019 (COVID-19)‒related research is essential, given the disease's novelty and its impact on society. However, with the publication pace speeding up, researchers and clinicians require automatic approaches to keep up with the incoming information regarding this disease. A solution to this problem requires the development of text mining pipelines; the efficiency of which strongly depends on the availability of curated corpora. However, there is a lack of COVID-19‒related corpora, even more, if considering other languages besides English. This project's main contribution was the annotation of a multilingual parallel corpus and the generation of a recommendation dataset (EN-PT and EN-ES) regarding relevant entities, their relations, and recommendation, providing this resource to the community to improve the text mining research on COVID-19‒related literature. This work was developed during the 7th Biomedical Linked Annotation Hackathon (BLAH7).

CT Angiography Clot Burden Score from Data Mining of Structured Reports for Pulmonary Embolism.

Background Many studies emphasize the role of structured reports (SRs) because they are readily accessible for further automated analyses. However, using SR data obtained in clinical routine for research purposes is not yet well represented in literature. Purpose To compare the performance of the Qanadli scoring system with a clot burden score mined from structured pulmonary embolism (PE) reports from CT angiography. Materials and Methods In this retrospective study, a rule-based text mining pipeline was developed to extract descriptors of PE and right heart strain from SR of patients with suspected PE between March 2017 and February 2020. From standardized PE reporting, a pulmonary artery obstruction index (PAOI) clot burden score (PAOICBS) was derived and compared with the Qanadli score (PAOIQ). Scoring time and confidence from two independent readings were compared. Interobserver and interscore agreement was tested by using the intraclass correlation coefficient (ICC) and Bland-Altman analysis. To assess conformity and diagnostic performance of both scores, areas under the receiver operating characteristic curve (AUCs) were calculated to predict right heart strain incidence, as were optimal cutoff values for maximum sensitivity and specificity. Results SR content authored by 67 residents and signed off by 32 consultants from 1248 patients (mean age, 63 years ± 17 [standard deviation]; 639 men) was extracted accurately and allowed for PAOICBS calculation in 304 of 357 (85.2%) PE-positive reports. The PAOICBS strongly correlated with the PAOIQ (r = 0.94; P < .001). Use of PAOICBS yielded overall time savings (1.3 minutes ± 0.5 vs 3.0 minutes ± 1.7), higher confidence levels (4.2 ± 0.6 vs 3.6 ± 1.0), and a higher ICC (ICC, 0.99 vs 0.95), respectively, compared with PAOIQ (each, P < .001). AUCs were similar for PAOICBS (AUC, 0.75; 95% CI: 0.70, 0.81) and PAOIQ (AUC, 0.77; 95% CI: 0.72, 0.83; P = .68), with cutoff values of 27.5% for both scores. Conclusion Data mining of structured reports enabled the development of a CT angiography scoring system that simplified the Qanadli score as a semiquantitative estimate of thrombus burden in patients with pulmonary embolism. © RSNA, 2021 Online supplemental material is available for this article. See also the editorial by Hunsaker in this issue.

Synthetic Biology Knowledge System.

The Synthetic Biology Knowledge System (SBKS) is an instance of the SynBioHub repository that includes text and data information that has been mined from papers published in ACS Synthetic Biology. This paper describes the SBKS curation framework that is being developed to construct the knowledge stored in this repository. The text mining pipeline performs automatic annotation of the articles using natural language processing techniques to identify salient content such as key terms, relationships between terms, and main topics. The data mining pipeline performs automatic annotation of the sequences extracted from the supplemental documents with the genetic parts used in them. Together these two pipelines link genetic parts to papers describing the context in which they are used. Ultimately, SBKS will reduce the time necessary for synthetic biologists to find the information necessary to complete their designs.

Text mining of gene\u2013phenotype associations reveals new phenotypic profiles of autism-associated genes

Autism is a spectrum disorder with wide variation in type and severity of symptoms. Understanding gene–phenotype associations is vital to unravel the disease mechanisms and advance its diagnosis and treatment. To date, several databases have stored a large portion of gene–phenotype associations which are mainly obtained from genetic experiments. However, a large proportion of gene–phenotype associations are still buried in the autism-related literature and there are limited resources to investigate autism-associated gene–phenotype associations. Given the abundance of the autism-related literature, we were thus motivated to develop Autism_genepheno, a text mining pipeline to identify sentence-level mentions of autism-associated genes and phenotypes in literature through natural language processing methods. We have generated a comprehensive database of gene–phenotype associations in the last five years’ autism-related literature that can be easily updated as new literature becomes available. We have evaluated our pipeline through several different approaches, and we are able to rank and select top autism-associated genes through their unique and wide spectrum of phenotypic profiles, which could provide a unique resource for the diagnosis and treatment of autism. The data resources and the Autism_genpheno pipeline are available at: https://github.com/maiziezhoulab/Autism_genepheno.

Automatic Prediction of Recurrence of Major Cardiovascular Events: A Text Mining Study Using Chest X-Ray Reports.

Methods We used EHR data of patients included in the Second Manifestations of ARTerial disease (SMART) study. We propose a deep learning-based multimodal architecture for our text mining pipeline that integrates neural text representation with preprocessed clinical predictors for the prediction of recurrence of major cardiovascular events in cardiovascular patients. Text preprocessing, including cleaning and stemming, was first applied to filter out the unwanted texts from X-ray radiology reports. Thereafter, text representation methods were used to numerically represent unstructured radiology reports with vectors. Subsequently, these text representation methods were added to prediction models to assess their clinical relevance. In this step, we applied logistic regression, support vector machine (SVM), multilayer perceptron neural network, convolutional neural network, long short-term memory (LSTM), and bidirectional LSTM deep neural network (BiLSTM). Results We performed various experiments to evaluate the added value of the text in the prediction of major cardiovascular events. The two main scenarios were the integration of radiology reports (1) with classical clinical predictors and (2) with only age and sex in the case of unavailable clinical predictors. In total, data of 5603 patients were used with 5-fold cross-validation to train the models. In the first scenario, the multimodal BiLSTM (MI-BiLSTM) model achieved an area under the curve (AUC) of 84.7%, misclassification rate of 14.3%, and F1 score of 83.8%. In this scenario, the SVM model, trained on clinical variables and bag-of-words representation, achieved the lowest misclassification rate of 12.2%. In the case of unavailable clinical predictors, the MI-BiLSTM model trained on radiology reports and demographic (age and sex) variables reached an AUC, F1 score, and misclassification rate of 74.5%, 70.8%, and 20.4%, respectively. Conclusions Using the case study of routine care chest X-ray radiology reports, we demonstrated the clinical relevance of integrating text features and classical predictors in our text mining pipeline for cardiovascular risk prediction. The MI-BiLSTM model with word embedding representation appeared to have a desirable performance when trained on text data integrated with the clinical variables from the SMART study. Our results mined from chest X-ray reports showed that models using text data in addition to laboratory values outperform those using only known clinical predictors.

A fast, accurate, and generalisable heuristic-based negation detection algorithm for clinical text

Negation detection is an important task in biomedical text mining. Particularly in clinical settings, it is of critical importance to determine whether findings mentioned in text are present or absent. Rule-based negation detection algorithms are a common approach to the task, and more recent investigations have resulted in the development of rule-based systems utilising the rich grammatical information afforded by typed dependency graphs. However, interacting with these complex representations inevitably necessitates complex rules, which are time-consuming to develop and do not generalise well. We hypothesise that a heuristic approach to determining negation via dependency graphs could offer a powerful alternative. We describe and implement an algorithm for negation detection based on grammatical distance from a negatory construct in a typed dependency graph. To evaluate the algorithm, we develop two testing corpora comprised of sentences of clinical text extracted from the MIMIC-III database and documents related to hypertrophic cardiomyopathy patients routinely collected at University Hospitals Birmingham NHS trust. Gold-standard validation datasets were built by a combination of human annotation and examination of algorithm error. Finally, we compare the performance of our approach with four other rule-based algorithms on both gold-standard corpora. The presented algorithm exhibits the best performance by f-measure over the MIMIC-III dataset, and a similar performance to the syntactic negation detection systems over the HCM dataset. It is also the fastest of the dependency-based negation systems explored in this study. Our results show that while a single heuristic approach to dependency-based negation detection is ignorant to certain advanced cases, it nevertheless forms a powerful and stable method, requiring minimal training and adaptation between datasets. As such, it could present a drop-in replacement or augmentation for many-rule negation approaches in clinical text-mining pipelines, particularly for cases where adaptation and rule development is not required or possible.

Comparison of rule-based and neural network models for negation detection in radiology reports

AbstractUsing natural language processing, it is possible to extract structured information from raw text in the electronic health record (EHR) at reasonably high accuracy. However, the accurate distinction between negated and non-negated mentions of clinical terms remains a challenge. EHR text includes cases where diseases are stated not to be present or only hypothesised, meaning a disease can be mentioned in a report when it is not being reported as present. This makes tasks such as document classification and summarisation more difficult. We have developed the rule-based EdIE-R-Neg, part of an existing text mining pipeline called EdIE-R (Edinburgh Information Extraction for Radiology reports), developed to process brain imaging reports, (https://www.ltg.ed.ac.uk/software/edie-r/) and two machine learning approaches; one using a bidirectional long short-term memory network and another using a feedforward neural network. These were developed on data from the Edinburgh Stroke Study (ESS) and tested on data from routine reports from NHS Tayside (Tayside). Both datasets consist of written reports from medical scans. These models are compared with two existing rule-based models: pyConText (Harkema et al. 2009. Journal of Biomedical Informatics42(5), 839–851), a python implementation of a generalisation of NegEx, and NegBio (Peng et al. 2017. NegBio: A high-performance tool for negation and uncertainty detection in radiology reports. arXiv e-prints, p. arXiv:1712.05898), which identifies negation scopes through patterns applied to a syntactic representation of the sentence. On both the test set of the dataset from which our models were developed, as well as the largely similar Tayside test set, the neural network models and our custom-built rule-based system outperformed the existing methods. EdIE-R-Neg scored highest on F1 score, particularly on the test set of the Tayside dataset, from which no development data were used in these experiments, showing the power of custom-built rule-based systems for negation detection on datasets of this size. The performance gap of the machine learning models to EdIE-R-Neg on the Tayside test set was reduced through adding development Tayside data into the ESS training set, demonstrating the adaptability of the neural network models.

Linking chemical and disease entities to ontologies by integrating PageRank with extracted relations from literature

BackgroundNamed Entity Linking systems are a powerful aid to the manual curation of digital libraries, which is getting increasingly costly and inefficient due to the information overload. Models based on the Personalized PageRank (PPR) algorithm are one of the state-of-the-art approaches, but these have low performance when the disambiguation graphs are sparse.FindingsThis work proposes a Named Entity Linking framework designated by Relation Extraction for Entity Linking (REEL) that uses automatically extracted relations to overcome this limitation. Our method builds a disambiguation graph, where the nodes are the ontology candidates for the entities and the edges are added according to the relations established in the text, which the method extracts automatically. The PPR algorithm and the information content of each ontology are then applied to choose the candidate for each entity that maximises the coherence of the disambiguation graph. We evaluated the method on three gold standards: the subset of the CRAFT corpus with ChEBI annotations (CRAFT-ChEBI), the subset of the BC5CDR corpus with disease annotations from the MEDIC vocabulary (BC5CDR-Diseases) and the subset with chemical annotations from the CTD-Chemical vocabulary (BC5CDR-Chemicals). The F1-Score achieved by REEL was 85.8%, 80.9% and 90.3% in these gold standards, respectively, outperforming baseline approaches.ConclusionsWe demonstrated that RE tools can improve Named Entity Linking by capturing semantic information expressed in text missing in Knowledge Bases and use it to improve the disambiguation graph of Named Entity Linking models. REEL can be adapted to any text mining pipeline and potentially to any domain, as long as there is an ontology or other knowledge Base available.