Dina Vishnyakova

The gene normalization task in BioCreative III

BackgroundWe report the Gene Normalization (GN) challenge in BioCreative III where participating teams were asked to return a ranked list of identifiers of the genes detected in full-text articles. For training, 32 fully and 500 partially annotated articles were prepared. A total of 507 articles were selected as the test set. Due to the high annotation cost, it was not feasible to obtain gold-standard human annotations for all test articles. Instead, we developed an Expectation Maximization (EM) algorithm approach for choosing a small number of test articles for manual annotation that were most capable of differentiating team performance. Moreover, the same algorithm was subsequently used for inferring ground truth based solely on team submissions. We report team performance on both gold standard and inferred ground truth using a newly proposed metric called Threshold Average Precision (TAP-k).ResultsWe received a total of 37 runs from 14 different teams for the task. When evaluated using the gold-standard annotations of the 50 articles, the highest TAP-k scores were 0.3297 (k=5), 0.3538 (k=10), and 0.3535 (k=20), respectively. Higher TAP-k scores of 0.4916 (k=5, 10, 20) were observed when evaluated using the inferred ground truth over the full test set. When combining team results using machine learning, the best composite system achieved TAP-k scores of 0.3707 (k=5), 0.4311 (k=10), and 0.4477 (k=20) on the gold standard, representing improvements of 12.4%, 21.8%, and 26.6% over the best team results, respectively.ConclusionsBy using full text and being species non-specific, the GN task in BioCreative III has moved closer to a real literature curation task than similar tasks in the past and presents additional challenges for the text mining community, as revealed in the overall team results. By evaluating teams using the gold standard, we show that the EM algorithm allows team submissions to be differentiated while keeping the manual annotation effort feasible. Using the inferred ground truth we show measures of comparative performance between teams. Finally, by comparing team rankings on gold standard vs. inferred ground truth, we further demonstrate that the inferred ground truth is as effective as the gold standard for detecting good team performance.

Managing the data deluge: data-driven GO category assignment improves while complexity of functional annotation increases

The available curated data lag behind current biological knowledge contained in the literature. Text mining can assist biologists and curators to locate and access this knowledge, for instance by characterizing the functional profile of publications. Gene Ontology (GO) category assignment in free text already supports various applications, such as powering ontology-based search engines, finding curation-relevant articles (triage) or helping the curator to identify and encode functions. Popular text mining tools for GO classification are based on so called thesaurus-based—or dictionary-based—approaches, which exploit similarities between the input text and GO terms themselves. But their effectiveness remains limited owing to the complex nature of GO terms, which rarely occur in text. In contrast, machine learning approaches exploit similarities between the input text and already curated instances contained in a knowledge base to infer a functional profile. GO Annotations (GOA) and MEDLINE make possible to exploit a growing amount of curated abstracts (97 000 in November 2012) for populating this knowledge base. Our study compares a state-of-the-art thesaurus-based system with a machine learning system (based on a k-Nearest Neighbours algorithm) for the task of proposing a functional profile for unseen MEDLINE abstracts, and shows how resources and performances have evolved. Systems are evaluated on their ability to propose for a given abstract the GO terms (2.8 on average) used for curation in GOA. We show that since 2006, although a massive effort was put into adding synonyms in GO (+300%), our thesaurus-based system effectiveness is rather constant, reaching from 0.28 to 0.31 for Recall at 20 (R20). In contrast, thanks to its knowledge base growth, our machine learning system has steadily improved, reaching from 0.38 in 2006 to 0.56 for R20 in 2012. Integrated in semi-automatic workflows or in fully automatic pipelines, such systems are more and more efficient to provide assistance to biologists. Database URL: http://eagl.unige.ch/GOCat/

Deep Question Answering for protein annotation

Biomedical professionals have access to a huge amount of literature, but when they use a search engine, they often have to deal with too many documents to efficiently find the appropriate information in a reasonable time. In this perspective, question-answering (QA) engines are designed to display answers, which were automatically extracted from the retrieved documents. Standard QA engines in literature process a user question, then retrieve relevant documents and finally extract some possible answers out of these documents using various named-entity recognition processes. In our study, we try to answer complex genomics questions, which can be adequately answered only using Gene Ontology (GO) concepts. Such complex answers cannot be found using state-of-the-art dictionary- and redundancy-based QA engines. We compare the effectiveness of two dictionary-based classifiers for extracting correct GO answers from a large set of 100 retrieved abstracts per question. In the same way, we also investigate the power of GOCat, a GO supervised classifier. GOCat exploits the GOA database to propose GO concepts that were annotated by curators for similar abstracts. This approach is called deep QA, as it adds an original classification step, and exploits curated biological data to infer answers, which are not explicitly mentioned in the retrieved documents. We show that for complex answers such as protein functional descriptions, the redundancy phenomenon has a limited effect. Similarly usual dictionary-based approaches are relatively ineffective. In contrast, we demonstrate how existing curated data, beyond information extraction, can be exploited by a supervised classifier, such as GOCat, to massively improve both the quantity and the quality of the answers with a +100% improvement for both recall and precision. Database URL: http://eagl.unige.ch/DeepQA4PA/

Using binary classification to prioritize and curate articles for the Comparative Toxicogenomics Database.

We report on the original integration of an automatic text categorization pipeline, so-called ToxiCat (Toxicogenomic Categorizer), that we developed to perform biomedical documents classification and prioritization in order to speed up the curation of the Comparative Toxicogenomics Database (CTD). The task can be basically described as a binary classification task, where a scoring function is used to rank a selected set of articles. Then components of a question-answering system are used to extract CTD-specific annotations from the ranked list of articles. The ranking function is generated using a Support Vector Machine, which combines three main modules: an information retrieval engine for MEDLINE (EAGLi), a gene normalization service (NormaGene) developed for a previous BioCreative campaign and finally, a set of answering components and entity recognizer for diseases and chemicals. The main components of the pipeline are publicly available both as web application and web services. The specific integration performed for the BioCreative competition is available via a web user interface at http://pingu.unige.ch:8080/Toxicat.

KART, a knowledge authoring and refinement tool for clinical guidelines development

Optimal antibiotic prescriptions rely on evidence-based clinical guidelines, but creating such guidelines requires a time-consuming systematic review of the literature. We aim at facilitating this process by proposing an innovative tool to extract antibiotic treatments from the literature.