Ian Lewin

Assessment of NER solutions against the first and second CALBC Silver Standard Corpus

BackgroundCompetitions in text mining have been used to measure the performance of automatic text processing solutions against a manually annotated gold standard corpus (GSC). The preparation of the GSC is time-consuming and costly and the final corpus consists at the most of a few thousand documents annotated with a limited set of semantic groups. To overcome these shortcomings, the CALBC project partners (PPs) have produced a large-scale annotated biomedical corpus with four different semantic groups through the harmonisation of annotations from automatic text mining solutions, the first version of the Silver Standard Corpus (SSC-I). The four semantic groups are chemical entities and drugs (CHED), genes and proteins (PRGE), diseases and disorders (DISO) and species (SPE). This corpus has been used for the First CALBC Challenge asking the participants to annotate the corpus with their text processing solutions.ResultsAll four PPs from the CALBC project and in addition, 12 challenge participants (CPs) contributed annotated data sets for an evaluation against the SSC-I. CPs could ignore the training data and deliver the annotations from their genuine annotation system, or could train a machine-learning approach on the provided pre-annotated data. In general, the performances of the annotation solutions were lower for entities from the categories CHED and PRGE in comparison to the identification of entities categorized as DISO and SPE. The best performance over all semantic groups were achieved from two annotation solutions that have been trained on the SSC-I.The data sets from participants were used to generate the harmonised Silver Standard Corpus II (SSC-II), if the participant did not make use of the annotated data set from the SSC-I for training purposes. The performances of the participants’ solutions were again measured against the SSC-II. The performances of the annotation solutions showed again better results for DISO and SPE in comparison to CHED and PRGE.ConclusionsThe SSC-I delivers a large set of annotations (1,121,705) for a large number of documents (100,000 Medline abstracts). The annotations cover four different semantic groups and are sufficiently homogeneous to be reproduced with a trained classifier leading to an average F-measure of 85%. Benchmarking the annotation solutions against the SSC-II leads to better performance for the CPs’ annotation solutions in comparison to the SSC-I.

Entity Recognition in Parallel Multi-lingual Biomedical Corpora: The CLEF-ER Laboratory Overview

The identification and normalisation of biomedical entities from the scientific literature has a long tradition and a number of challenges have contributed to the development of reliable solutions. Increasingly patient records are processed to align their content with other biomedical data resources, but this approach requires analysing documents in different languages across Europe [1,2]. The CLEF-ER challenge has been organized by the Mantra project partners to improve entity recognition ER in multilingual documents. Several corpora in different languages, i.e. Medline titles, EMEA documents and patent claims, have been prepared to enable ER in parallel documents. The participants have been ask to annotate entity mentions with concept unique identifiers CUIs in the documents of their preferred non-English language. The evaluation determines the number of correctly identified entity mentions against a silver standard Task A and the performance measures for the identification of CUIs in the non-English corpora. The participants could make use of the prepared terminological resources for entity normalisation and of the English silver standard corpora SSCs as input for concept candidates in the non-English documents. The participants used different approaches including translation techniques and word or phrase alignments apart from lexical lookup and other text mining techniques. The performances for task A and B was lower for the patent corpus in comparison to Medline titles and EMEA documents. In the patent documents, chemical entities were identified at higher performance, whereas the other two document types cover a higher portion of medical terms. The number of novel terms provided from all corpora is currently under investigation. Altogether, the CLEF-ER challenge demonstrates the performances of annotation solutions in different languages against an SSC.

Harmonization of gene/protein annotations

MOTIVATION The recognition of named entities (NER) is an elementary task in biomedical text mining. A number of NER solutions have been proposed in recent years, taking advantage of available annotated corpora, terminological resources and machine-learning techniques. Currently, the best performing solutions combine the outputs from selected annotation solutions measured against a single corpus. However, little effort has been spent on a systematic analysis of methods harmonizing the annotation results and measuring against a combination of Gold Standard Corpora (GSCs). RESULTS We present Totum, a machine learning solution that harmonizes gene/protein annotations provided by heterogeneous NER solutions. It has been optimized and measured against a combination of manually curated GSCs. The performed experiments show that our approach improves the F-measure of state-of-the-art solutions by up to 10% (achieving ≈70%) in exact alignment and 22% (achieving ≈82%) in nested alignment. We demonstrate that our solution delivers reliable annotation results across the GSCs and it is an important contribution towards a homogeneous annotation of MEDLINE abstracts. AVAILABILITY AND IMPLEMENTATION Totum is implemented in Java and its resources are available at http://bioinformatics.ua.pt/totum

Evaluating gold standard corpora against gene/protein tagging solutions and lexical resources.

MotivationThe identification of protein and gene names (PGNs) from the scientific literature requires semantic resources: Terminological and lexical resources deliver the term candidates into PGN tagging solutions and the gold standard corpora (GSC) train them to identify term parameters and contextual features. Ideally all three resources, i.e. corpora, lexica and taggers, cover the same domain knowledge, and thus support identification of the same types of PGNs and cover all of them. Unfortunately, none of the three serves as a predominant standard and for this reason it is worth exploring, how these three resources comply with each other. We systematically compare different PGN taggers against publicly available corpora and analyze the impact of the included lexical resource in their performance. In particular, we determine the performance gains through false positive filtering, which contributes to the disambiguation of identified PGNs.ResultsIn general, machine learning approaches (ML-Tag) for PGN tagging show higher F1-measure performance against the BioCreative-II and Jnlpba GSCs (exact matching), whereas the lexicon based approaches (LexTag) in combination with disambiguation methods show better results on FsuPrge and PennBio. The ML-Tag solutions balance precision and recall, whereas the LexTag solutions have different precision and recall profiles at the same F1-measure across all corpora. Higher recall is achieved with larger lexical resources, which also introduce more noise (false positive results). The ML-Tag solutions certainly perform best, if the test corpus is from the same GSC as the training corpus. As expected, the false negative errors characterize the test corpora and – on the other hand – the profiles of the false positive mistakes characterize the tagging solutions. Lex-Tag solutions that are based on a large terminological resource in combination with false positive filtering produce better results, which, in addition, provide concept identifiers from a knowledge source in contrast to ML-Tag solutions.ConclusionThe standard ML-Tag solutions achieve high performance, but not across all corpora, and thus should be trained using several different corpora to reduce possible biases. The LexTag solutions have different profiles for their precision and recall performance, but with similar F1-measure. This result is surprising and suggests that they cover a portion of the most common naming standards, but cope differently with the term variability across the corpora. The false positive filtering applied to LexTag solutions does improve the results by increasing their precision without compromising significantly their recall. The harmonisation of the annotation schemes in combination with standardized lexical resources in the tagging solutions will enable their comparability and will pave the way for a shared standard.

Evaluation and cross-comparison of lexical entities of biological interest (LexEBI).

Motivation Biomedical entities, their identifiers and names, are essential in the representation of biomedical facts and knowledge. In the same way, the complete set of biomedical and chemical terms, i.e. the biomedical “term space” (the “Lexeome”), forms a key resource to achieve the full integration of the scientific literature with biomedical data resources: any identified named entity can immediately be normalized to the correct database entry. This goal does not only require that we are aware of all existing terms, but would also profit from knowing all their senses and their semantic interpretation (ambiguities, nestedness). Result This study compiles a resource for lexical terms of biomedical interest in a standard format (called “LexEBI”), determines the overall number of terms, their reuse in different resources and the nestedness of terms. LexEBI comprises references for protein and gene entries and their term variants and chemical entities amongst other terms. In addition, disease terms have been identified from Medline and PubmedCentral and added to LexEBI. Our analysis demonstrates that the baseforms of terms from the different semantic types show only little polysemous use. Nonetheless, the term variants of protein and gene names (PGNs) frequently contain species mentions, which should have been avoided according to protein annotation guidelines. Furthermore, the protein and gene entities as well as the chemical entities, both do comprise enzymes leading to hierarchical polysemy, and a large portion of PGNs make reference to a chemical entity. Altogether, according to our analysis based on the Medline distribution, 401,869 unique PGNs in the documents contain a reference to 25,022 chemical entities, 3,125 disease terms or 1,576 species mentions. Conclusion LexEBI delivers the complete biomedical and chemical Lexeome in a standardized representation (http://www.ebi.ac.uk/Rebholz-srv/LexEBI/). The resource provides the disease terms as open source content, and fully interlinks terms across resources.

Monitoring named entity recognition: the League Table

BackgroundNamed entity recognition (NER) is an essential step in automatic text processing pipelines. A number of solutions have been presented and evaluated against gold standard corpora (GSC). The benchmarking against GSCs is crucial, but left to the individual researcher. Herewith we present a League Table web site, which benchmarks NER solutions against selected public GSCs, maintains a ranked list and archives the annotated corpus for future comparisons.ResultsThe web site enables access to the different GSCs in a standardized format (IeXML). Upon submission of the annotated corpus the user has to describe the specification of the used solution and then uploads the annotated corpus for evaluation. The performance of the system is measured against one or more GSCs and the results are then added to the web site (“League Table”). It displays currently the results from publicly available NER solutions from the Whatizit infrastructure for future comparisons.ConclusionThe League Table enables the evaluation of NER solutions in a standardized infrastructure and monitors the results long-term. For access please go to http://wwwdev.ebi.ac.uk/Rebholz-srv/calbc/assessmentGSC/. Contact: rebholz@ifi.uzh.ch.

Exploitation of Cross-References between Terminological Resources within the CALBC Context

In this paper we have analyzed current terminological resources in order to extract available cross-references (i.e. mappings) between biomedical terminologies. These cross-references have been applied within the CALBC challenge, an initiative which aims at automatically creating a large-scale and community-wide text corpus annotated with biomedical entities. We also show that cross-references are essential for a precise evaluation of the identifiers associated with the annotated entities by the CALBC systems.