Large-scale directional relationship extraction and resolution

Comparing neural models for nested and overlapping biomedical event detection

Background

Nested and overlapping events are particularly frequent and informative structures in biomedical event extraction. However, state-of-the-art neural models either neglect those structures during learning or use syntactic features and external tools to detect them. To overcome these limitations, this paper presents and compares two neural models: a novel EXhaustive Neural Network (EXNN) and a Search-Based Neural Network (SBNN) for detection of nested and overlapping events.

Results

We evaluate the proposed models as an event detection component in isolation and within a pipeline setting. Evaluation in several annotated biomedical event extraction datasets shows that both EXNN and SBNN achieve higher performance in detecting nested and overlapping events, compared to the state-of-the-art model Turku Event Extraction System (TEES).

Conclusions

The experimental results reveal that both EXNN and SBNN are effective for biomedical event extraction. Furthermore, results on a pipeline setting indicate that our models improve detection of events compared to models that use either gold or predicted named entities.

Automatic extraction of protein-protein interactions using grammatical relationship graph

Background

Relationships between bio-entities (genes, proteins, diseases, etc.) constitute a significant part of our knowledge. Most of this information is documented as unstructured text in different forms, such as books, articles and on-line pages. Automatic extraction of such information and storing it in structured form could help researchers more easily access such information and also make it possible to incorporate it in advanced integrative analysis. In this study, we developed a novel approach to extract bio-entity relationships information using Nature Language Processing (NLP) and a graph-theoretic algorithm.

Methods

Our method, called GRGT (Grammatical Relationship Graph for Triplets), not only extracts the pairs of terms that have certain relationships, but also extracts the type of relationship (the word describing the relationships). In addition, the directionality of the relationship can also be extracted. Our method is based on the assumption that a triplet exists for a pair of interactions. A triplet is defined as two terms (entities) and an interaction word describing the relationship of the two terms in a sentence. We first use a sentence parsing tool to obtain the sentence structure represented as a dependency graph where words are nodes and edges are typed dependencies. The shortest paths among the pairs of words in the triplet are then extracted, which form the basis for our information extraction method. Flexible pattern matching scheme was then used to match a triplet graph with unknown relationship to those triplet graphs with labels (True or False) in the database.

Results

We applied the method on three benchmark datasets to extract the protein-protein-interactions (PPIs), and obtained better precision than the top performing methods in literature.

Conclusions

We have developed a method to extract the protein-protein interactions from biomedical literature. PPIs extracted by our method have higher precision among other methods, suggesting that our method can be used to effectively extract PPIs and deposit them into databases. Beyond extracting PPIs, our method could be easily extended to extracting relationship information between other bio-entities.

ELTD1, an effective anti-angiogenic target for gliomas: preclinical assessment in mouse GL261 and human G55 xenograft glioma models

Background

Despite current therapies, glioblastoma is a devastating cancer, and validation of effective biomarkers for it will enable better diagnosis and therapeutic intervention for this disease. We recently discovered a new biomarker for high-grade gliomas, ELTD1 (epidermal growth factor, latrophilin, and 7 transmembrane domain-containing protein 1 on chromosome 1) via bioinformatics, and validated that ELTD1 protein levels are significantly higher in human and rodent gliomas. The focus of this study was to assess the effect on tumor growth of an antibody against ELTD1 in orthotopic, GL261, and G55 xenograft glioma models.

Methods

The effect of anti-ELTD1 antibody therapy was assessed by animal survival, MRI measured tumor volumes, MR angiography, MR perfusion imaging, and immunohistochemistry (IHC) characterization of microvessel density in mouse glioma models. Comparative treatments included anti-vascular endothelial growth factor (VEGF) and anti-c-Met antibody therapies, compared with untreated controls.

Results

Tumor volume and survival data in this study show that antibodies against ELTD1 inhibit glioma growth just as effectively or even more so compared with other therapeutic targets studied, including anti-VEGF antibody therapy. Untreated GL261 or G55 tumors were found to have significantly higher ELTD1 levels (IHC) compared with contralateral normal brain. The anti-angiogenic effect of ELTD1 antibody therapy was observed in assessment of microvessel density, as well as from MR angiography and perfusion measurements, which indicated that anti-ELTD1 antibody therapy significantly decreased vascularization compared with untreated controls.

Conclusions

Either as a single therapy or in conjunction with other therapeutic approaches, anti-ELTD1 antibodies could be a valuable new clinical anti-angiogenic therapeutic for high-grade gliomas.

Knowledge Extraction from MEDLINE by Combining Clustering with Natural Language Processing

The identification of relevant predicates between co-occurring concepts in scientific literature databases like MEDLINE is crucial for using these sources for knowledge extraction, in order to obtain meaningful biomedical predications as subject-predicate-object triples. We consider the manually assigned MeSH indexing terms (main headings and subheadings) in MEDLINE records as a rich resource for extracting a broad range of domain knowledge. In this paper, we explore the combination of a clustering method for co-occurring concepts based on their related MeSH subheadings in MEDLINE with the use of SemRep, a natural language processing engine, which extracts predications from free text documents. As a result, we generated sets of clusters of co-occurring concepts and identified the most significant predicates for each cluster. The association of such predicates with the co-occurrences of the resulting clusters produces the list of predications, which were checked for relevance.

Experimental validation of 5 in-silico predicted glioma biomarkers

BACKGROUND

Glioblastoma multiforme (GBM) is a high-grade glioma with poor prognosis. Identification of new biomarkers specific to GBM could help in disease diagnosis. We have developed and validated a bioinformatics method to predict proteins likely to be suitable as glioma biomarkers via a global microarray meta-analysis to identify uncharacterized genes consistently coexpressed with known glioma-associated genes.

METHODS

A novel bioinformatics method was implemented called global microarray meta-analysis, using approximately 16,000 microarray experiments to identify uncharacterized genes consistently coexpressed with known glioma-associated genes. These novel biomarkers were validated as proteins highly expressed in human gliomas varying in tumor grades using immunohistochemistry. Glioma gene databases were used to assess delineation of expression of these markers in varying glioma grades and subtypes of GBM.

RESULTS

We have identified 5 potential biomarkers-spondin1, Plexin-B2, SLIT3, fibulin-1, and LINGO1-that were validated as proteins highly expressed on the surface of human gliomas using immunohistochemistry. Expression of spondin1, Plexin-B2, and SLIT3 was significantly higher (P < .01) in high-grade gliomas than in low-grade gliomas. These biomarkers were significant discriminators in grade IV gliomas compared with either grade III or II tumors and also distinguished between GBM subclasses.

CONCLUSIONS

This study strongly suggests that this type of bioinformatics approach has high translational potential to rapidly discern which poorly characterized proteins may be of clinical relevance.

ELTD1, a potential new biomarker for gliomas

BACKGROUND

Glioblastoma multiforme (GBM), a high-grade glioma, is characterized by being diffuse, invasive, and highly angiogenic and has a very poor prognosis. Identification of new biomarkers could help in the further diagnosis of GBM.

OBJECTIVE

To identify ELTD1 (epidermal growth factor, latrophilin, and 7 transmembrane domain-containing protein 1 on chromosome 1) as a putative glioma-associated marker via a bioinformatic method.

METHODS

We used advanced data mining and a novel bioinformatics method to predict ELTD1 as a potential novel biomarker that is associated with gliomas. Validation was done with immunohistochemistry, which was used to detect levels of ELTD1 in human high-grade gliomas and rat F98 glioma tumors. In vivo levels of ELTD1 in rat F98 gliomas were assessed using molecular magnetic resonance imaging.

RESULTS

ELTD1 was found to be significantly higher (P = .03) in high-grade gliomas (50 patients) compared with low-grade gliomas (21 patients) and compared well with traditional immunohistochemistry markers including vascular endothelial growth factor, glucose transporter 1, carbonic anhydrase IX, and hypoxia-inducible factor 1α. ELTD1 gene expression indicates an association with grade, survival across grade, and an increase in the mesenchymal subtype. Significantly high (P < .001) in vivo levels of ELTD1 were additionally found in F98 tumors compared with normal brain tissue.

CONCLUSION

Results of this study strongly suggests that associative analysis was able to accurately identify ELTD1 as a putative glioma-associated biomarker. The detection of ELTD1 was also validated in both rodent and human gliomas and may serve as an additional biomarker for gliomas in preclinical and clinical diagnosis of gliomas.

PPInterFinder--a mining tool for extracting causal relations on human proteins from literature

One of the most common and challenging problem in biomedical text mining is to mine protein-protein interactions (PPIs) from MEDLINE abstracts and full-text research articles because PPIs play a major role in understanding the various biological processes and the impact of proteins in diseases. We implemented, PPInterFinder--a web-based text mining tool to extract human PPIs from biomedical literature. PPInterFinder uses relation keyword co-occurrences with protein names to extract information on PPIs from MEDLINE abstracts and consists of three phases. First, it identifies the relation keyword using a parser with Tregex and a relation keyword dictionary. Next, it automatically identifies the candidate PPI pairs with a set of rules related to PPI recognition. Finally, it extracts the relations by matching the sentence with a set of 11 specific patterns based on the syntactic nature of PPI pair. We find that PPInterFinder is capable of predicting PPIs with the accuracy of 66.05% on AIMED corpus and outperforms most of the existing systems. DATABASE URL: http://www.biomining-bu.in/ppinterfinder/

IMID: integrated molecular interaction database

MOTIVATION

Molecular interaction information, such as protein-protein interactions and protein-small molecule interactions, is indispensable for understanding the mechanism of biological processes and discovering treatments for diseases. Many databases have been built by manual annotation of literature to organize such information into structured form. However, most databases focus on only one type of interactions, which are often not well annotated and integrated with related functional information.

RESULTS

In this study, we integrate molecular interaction information from literature by automatic information extraction and from manually annotated databases. We further integrate the relationships between protein/gene and other bio-entity terms including gene ontology terms, pathways, species and diseases to build an integrated molecular interaction database (IMID). Interactions can be selected by their associated probabilities. IMID allows complex and versatile queries for context-specific molecular interactions, which are not available currently in other molecular interaction databases.

AVAILABILITY

The database is located at www.integrativebiology.org.

Integrated bio-entity network: a system for biological knowledge discovery

A significant part of our biological knowledge is centered on relationships between biological entities (bio-entities) such as proteins, genes, small molecules, pathways, gene ontology (GO) terms and diseases. Accumulated at an increasing speed, the information on bio-entity relationships is archived in different forms at scattered places. Most of such information is buried in scientific literature as unstructured text. Organizing heterogeneous information in a structured form not only facilitates study of biological systems using integrative approaches, but also allows discovery of new knowledge in an automatic and systematic way. In this study, we performed a large scale integration of bio-entity relationship information from both databases containing manually annotated, structured information and automatic information extraction of unstructured text in scientific literature. The relationship information we integrated in this study includes protein–protein interactions, protein/gene regulations, protein–small molecule interactions, protein–GO relationships, protein–pathway relationships, and pathway–disease relationships. The relationship information is organized in a graph data structure, named integrated bio-entity network (IBN), where the vertices are the bio-entities and edges represent their relationships. Under this framework, graph theoretic algorithms can be designed to perform various knowledge discovery tasks. We designed breadth-first search with pruning (BFSP) and most probable path (MPP) algorithms to automatically generate hypotheses—the indirect relationships with high probabilities in the network. We show that IBN can be used to generate plausible hypotheses, which not only help to better understand the complex interactions in biological systems, but also provide guidance for experimental designs.

BICEPP: an example-based statistical text mining method for predicting the binary characteristics of drugs

Background

The identification of drug characteristics is a clinically important task, but it requires much expert knowledge and consumes substantial resources. We have developed a statistical text-mining approach (BInary Characteristics Extractor and biomedical Properties Predictor: BICEPP) to help experts screen drugs that may have important clinical characteristics of interest.

Results

BICEPP first retrieves MEDLINE abstracts containing drug names, then selects tokens that best predict the list of drugs which represents the characteristic of interest. Machine learning is then used to classify drugs using a document frequency-based measure. Evaluation experiments were performed to validate BICEPP's performance on 484 characteristics of 857 drugs, identified from the Australian Medicines Handbook (AMH) and the PharmacoKinetic Interaction Screening (PKIS) database. Stratified cross-validations revealed that BICEPP was able to classify drugs into all 20 major therapeutic classes (100%) and 157 (of 197) minor drug classes (80%) with areas under the receiver operating characteristic curve (AUC) > 0.80. Similarly, AUC > 0.80 could be obtained in the classification of 173 (of 238) adverse events (73%), up to 12 (of 15) groups of clinically significant cytochrome P450 enzyme (CYP) inducers or inhibitors (80%), and up to 11 (of 14) groups of narrow therapeutic index drugs (79%). Interestingly, it was observed that the keywords used to describe a drug characteristic were not necessarily the most predictive ones for the classification task.

Conclusions

BICEPP has sufficient classification power to automatically distinguish a wide range of clinical properties of drugs. This may be used in pharmacovigilance applications to assist with rapid screening of large drug databases to identify important characteristics for further evaluation.

A comprehensive benchmark of kernel methods to extract protein-protein interactions from literature

The most important way of conveying new findings in biomedical research is scientific publication. Extraction of protein-protein interactions (PPIs) reported in scientific publications is one of the core topics of text mining in the life sciences. Recently, a new class of such methods has been proposed - convolution kernels that identify PPIs using deep parses of sentences. However, comparing published results of different PPI extraction methods is impossible due to the use of different evaluation corpora, different evaluation metrics, different tuning procedures, etc. In this paper, we study whether the reported performance metrics are robust across different corpora and learning settings and whether the use of deep parsing actually leads to an increase in extraction quality. Our ultimate goal is to identify the one method that performs best in real-life scenarios, where information extraction is performed on unseen text and not on specifically prepared evaluation data. We performed a comprehensive benchmarking of nine different methods for PPI extraction that use convolution kernels on rich linguistic information. Methods were evaluated on five different public corpora using cross-validation, cross-learning, and cross-corpus evaluation. Our study confirms that kernels using dependency trees generally outperform kernels based on syntax trees. However, our study also shows that only the best kernel methods can compete with a simple rule-based approach when the evaluation prevents information leakage between training and test corpora. Our results further reveal that the F-score of many approaches drops significantly if no corpus-specific parameter optimization is applied and that methods reaching a good AUC score often perform much worse in terms of F-score. We conclude that for most kernels no sensible estimation of PPI extraction performance on new text is possible, given the current heterogeneity in evaluation data. Nevertheless, our study shows that three kernels are clearly superior to the other methods.

Extracting causal relations on HIV drug resistance from literature

BACKGROUND

In HIV treatment it is critical to have up-to-date resistance data of applicable drugs since HIV has a very high rate of mutation. These data are made available through scientific publications and must be extracted manually by experts in order to be used by virologists and medical doctors. Therefore there is an urgent need for a tool that partially automates this process and is able to retrieve relations between drugs and virus mutations from literature.

RESULTS

In this work we present a novel method to extract and combine relationships between HIV drugs and mutations in viral genomes. Our extraction method is based on natural language processing (NLP) which produces grammatical relations and applies a set of rules to these relations. We applied our method to a relevant set of PubMed abstracts and obtained 2,434 extracted relations with an estimated performance of 84% for F-score. We then combined the extracted relations using logistic regression to generate resistance values for each <drug, mutation> pair. The results of this relation combination show more than 85% agreement with the Stanford HIVDB for the ten most frequently occurring mutations. The system is used in 5 hospitals from the Virolab project http://www.virolab.org to preselect the most relevant novel resistance data from literature and present those to virologists and medical doctors for further evaluation.

CONCLUSIONS

The proposed relation extraction and combination method has a good performance on extracting HIV drug resistance data. It can be used in large-scale relation extraction experiments. The developed methods can also be applied to extract other type of relations such as gene-protein, gene-disease, and disease-mutation.

PathBinder--text empirics and automatic extraction of biomolecular interactions

Motivation

The increasingly large amount of free, online biological text makes automatic interaction extraction correspondingly attractive. Machine learning is one strategy that works by uncovering and using useful properties that are implicit in the text. However these properties are usually not reported in the literature explicitly. By investigating specific properties of biological text passages in this paper, we aim to facilitate an alternative strategy, the use of text empirics, to support mining of biomedical texts for biomolecular interactions. We report on our application of this approach, and also report some empirical findings about an important class of passages. These may be useful to others who may also wish to use the empirical properties we describe.

Results

We manually analyzed syntactic and semantic properties of sentences likely to describe interactions between biomolecules. The resulting empirical data were used to design an algorithm for the PathBinder system to extract biomolecular interactions from texts. PathBinder searches PubMed for sentences describing interactions between two given biomolecules. PathBinder then uses probabilistic methods to combine evidence from multiple relevant sentences in PubMed to assess the relative likelihood of interaction between two arbitrary biomolecules. A biomolecular interaction network was constructed based on those likelihoods.

Conclusion

The text empirics approach used here supports computationally friendly, performance competitive, automatic extraction of biomolecular interactions from texts.

Availability

http://www.metnetdb.org/pathbinder.