Search

PROBLEM: key word assignment has been largely used in MEDLINE to provide an indicative “gist” of the content of articles and to help retrieving biomedical articles. Abstracts are also used for this purpose. However with usually more than 300 words, MEDLINE abstracts can still be regarded as long documents; therefore we design a system to select a unique key sentence. This key sentence must be indicative of the article's content and we assume that abstract's conclusions are good candidates. We design and assess the performance of an automatic key sentence selector, which classifies sentences into four argumentative moves: PURPOSE, METHODS, RESULTS and CONCLUSION. METHODS: we rely on Bayesian classifiers trained on automatically acquired data. Features representation, selection and weighting are reported and classification effectiveness is evaluated on the four classes using confusion matrices. We also explore the use of simple heuristics to take the position of sentences into account. Recall, precision and F-scores are computed for the CONCLUSION class. For the CONCLUSION class, the F-score reaches 84%. Automatic argumentative classification using Bayesian learners is feasible on MEDLINE abstracts and should help user navigation in such repositories.

N-terminal myristoylation is a post-translational modification that causes the addition of a myristate to a glycine in the N-terminal end of the amino acid chain. This work presents neural network (NN) models that learn to discriminate myristoylated and nonmyristoylated proteins. Ensembles of 25 NNs and decision trees were trained on 390 positive sequences and 327 negative sequences. Experiments showed that NN ensembles were more accurate than decision tree ensembles. Our NN predictor evaluated by the leave-one-out procedure, obtained a false positive error rate equal to 2.1%. That was better than the PROSITE pattern for myristoylation for which the false positive error rate was 22.3%. On a recent version of Swiss-Prot (41.2), the NN ensemble predicted 876 myristoylated proteins, while 1150 proteins were predicted by the PROSITE pattern for myristoylation. Finally, compared to the well-known NMT predictor, the NN predictor gave similar results. Our tool is available under http://www.expasy.org/tools/myristoylator/myristoylator.html.

Motivation: Searching relevant publications for manual database annotation is a tedious task. In this paper, we apply a combination of Natural Language Processing (NLP) and probabilistic classification to re-rank documents returned by PubMed according to their relevance to Swiss-Prot annotation, and to identify significant terms in the documents. Results: With a Probabilistic Latent Categoriser (PLC) we obtained 69% recall and 59% precision for relevant documents in a representative query. As the PLC technique provides the relative contribution of each term to the final document score, we used the Kullback-Leibler symmetric divergence to determine the most discriminating words for Swiss-Prot medical annotation. This information should allow curators to understand classification results better. It also has great value for fine-tuning the linguistic pre-processing of documents, which in turn can improve the overall classifier performance. Availability: The medical annotation dataset is available from the authors upon request Contact: Pavel.Dobrokhotov@isb-sib.ch; Cyril.Goutte@xrce.xerox.com *To whom correspondence should be addressed.

Motivation: Post-translational modifications (PTMs) are important steps in the maturation of proteins. Several models exist to predict specific PTMs, from manually detected patterns to machine learning methods. On one hand, the manual detection of patterns does not provide the most efficient classifiers and requires an important workload, and on the other hand, models built by machine learning methods are hard to interpret and do not increase biological knowledge. Therefore, we developed a novel method based on patterns discovery and decision trees to predict PTMs. The proposed algorithm builds a decision tree, by coupling the C4.5 algorithm with genetic algorithms, producing high-performance white box classifiers. Our method was tested on the initiator methionine cleavage (IMC) and N α -terminal acetylation (N-Ac), two of the most common PTMs. Results: The resulting classifiers perform well when compared with existing models. On a set of eukaryotic proteins, they display a cross-validated Matthews correlation coefficient of 0.83 (IMC) and 0.65 (N-Ac). When used to predict potential substrates of N-terminal acetyltransferaseB and N-terminal acetyltransferaseC, our classifiers display better performance than the state of the art. Moreover, we present an analysis of the model predicting IMC for Homo sapiens proteins and demonstrate that we are able to extract experimentally known facts without prior knowledge. Those results validate the fact that our method produces white box models. Availability and implementation: Predictors for IMC and N-Ac and all datasets are freely available at http://terminus.unige.ch/ . Contact: jean-luc.falcone@unige.ch Supplementary information: Supplementary data are available at Bioinformatics online.

Molecular biology has become heavily dependent on biological knowledge encoded in expert curated biological databases. As the volume of biological literature increases, biocurators need help in keeping up with the literature; (semi-) automated aids for biocuration would seem to be an ideal application for natural language processing and text mining. However, to date, there have been few documented successes for improving biocuration throughput using text mining. Our initial investigations took place for the workshop on ‘Text Mining for the BioCuration Workflow’ at the third International Biocuration Conference (Berlin, 2009). We interviewed biocurators to obtain workflows from eight biological databases. This initial study revealed high-level commonalities, including (i) selection of documents for curation; (ii) indexing of documents with biologically relevant entities (e.g. genes); and (iii) detailed curation of specific relations (e.g. interactions); however, the detailed workflows also showed many variabilities. Following the workshop, we conducted a survey of biocurators. The survey identified biocurator priorities, including the handling of full text indexed with biological entities and support for the identification and prioritization of documents for curation. It also indicated that two-thirds of the biocuration teams had experimented with text mining and almost half were using text mining at that time. Analysis of our interviews and survey provide a set of requirements for the integration of text mining into the biocuration workflow. These can guide the identification of common needs across curated databases and encourage joint experimentation involving biocurators, text mining developers and the larger biomedical research community., National Science Foundation (grant IIS-0844419 to L.H.); US National Institutes of Health National Library of Medicine (grant 1G08LM10720-01 to C.N.A. and C. H. W.); Work related to BioCreative III was supported by the US National Science Foundation (grant DBI-0850319 to C.N.A., L.H., C.H.W.); the US National Institute of General Medical Sciences (grant R01-GM083871 to G.A.P.C.B.); the National Science Foundation (DBI-0849977 to G.A.P.G.B); the European Union Seventh Framework MICROME project (Grant Agreement Number 222886-2 to M.K. and A.V.); the US National Science Foundation IGERT (Grant 0221625 to K.G.D) and a PhRMA Foundation predoctoral fellowship in informatics; US National Science Foundation (grant DBI-0850219 to E.H.); US National Human Genome Research Institute (grant HG001315 to R.N.); National Institutes of Health (NIH) (grant 2U01HG02712-04 to A.L.V.) and European Commission contract FELICS (grant 021902RII3); National Institute of Environmental Health Sciences (NIEHS) and the National Library of Medicine (NLM) (R01ES014065 to T.W.); NIEHS (R01ES014065-04S1 to T.W.); National Institutes of Health National Center for Research Resources(P20RR016463 to T.W.); Biotechnology and Biological Sciences Research Council of the UK (grant BB/F010486/1 to A.G.W); the National Institutes of Health National Center for Research Resources (1R01RR024031 to A.G.W); the European Commission FP7 Program (2007223411 to A.G.W). Funding for open access charge: The MITRE Corporation.

In the human proteome, about 5'000 proteins lack experimentally validated functional information. In this work we propose to tackle the problem of human protein function prediction by three distinct supervised learning schemes: one-versus-all classification; tournament learning; multi-label learning. Target values of supervised learning models are represented by the nodes of a subset of the Gene Ontology, which is widely used as a benchmark for functional prediction. With an independent dataset including very difficult cases the recall measure reached a reasonable performance for the first 50 ranked predictions, on average; however, average precision was quite low.

A text mining group has been set up at the Swiss Institute of Bioinformatics, with objective to develop and adapt information retrieval and extraction tools to help Swiss-Prot curators in their daily annotation work. After over 7 year activities, this group has gathered a significant amount of experience about the need in text mining for biocuration. The first observation we made is that there is no “in-a-box” solution which can satisfy every needs. Each curator has his/her own strategy to find information from the literature and none of the existing information retrieval systems is able to compete with it, more for reason of habits than for reason of performance. Second observation: to be completely operative, an information retrieval system should be embedded in the annotation platform. For instance, it should be possible to copy/paste information, such as the article reference or some interesting sentences, directly in the database format. Most of the existing online programs are hardly adaptable for this task and their use usually results in additional editing efforts for the curators. From this observation, we can derive the fact that integrating text mining services is usually more costly than expected since wrappers and user interfaces need significant developments sometimes fairly user-specific. After noticing these problems in the design and use of a generic information retrieval system for the Swiss-Prot curators, we focused our effort on text mining applications for database update. The follow-up of the literature is essential in the process of database maintenance and there are needs for automatic information extraction tools on a large panel of topics. We developed several IE applications in the field of: - PTM information (phosphorylation, glycosylation, disulfide bridge) - Subcellular localization - Variant/mutation detection and characterization - New sequence with enzymatic activities - New characterization of enzymes. These tools are integrated into pipelines which follow PubMed daily outcomes and generate list of selected abstracts with highlights on the relevant sentences. These procedures are done independently of the usual annotation workflow, so that curators can mine these preselected data whenever they work on database entry updates. To conclude, we have identified big challenges in text mining services after discussion with the curators. One of them is the detection of novel information, especially those related to a new function or a new characterization of a protein or one of its close homologues. We are currently working on this task in the framework of the collaborative project “EAGL”. Another challenge is definitely the large-scale screening of newly published full-text papers to complement the often incomplete information in abstracts. This becomes more and more indispensable, not really for the annotation of widely studied “hot” proteins, but to find new data on uncharacterized ones. For instance, when no gene name has been attributed to a sequence, the only way to retrieve information is to use the orf names, which are never provided in abstracts. Finally, one should definitely stress that many of these information retrieval and extraction tasks could be greatly simplified with the requirement of metadata at the article submission time, such as an official HGNC gene name or a UniProt reference.

Large-scale sequencing of prokaryotic genomes demands the automation of certain annotation tasks currently manually performed in the production of the SWISS-PROT protein knowledgebase. The HAMAP project, or 'High-quality Automated and Manual Annotation of microbial Proteomes', aims to integrate manual and automatic annotation methods in order to enhance the speed of the curation process while preserving the quality of the database annotation. Automatic annotation is only applied to entries that belong to manually defined orthologous families and to entries with no identifiable similarities (ORFans). Many checks are enforced in order to prevent the propagation of wrong annotation and to spot problematic cases, which are channelled to manual curation. The results of this annotation are integrated in SWISS-PROT, and a website is provided at http://www.expasy.org/sprot/hamap/.

Motivation and Objectives Biomedical professionals have at their disposal a huge amount of literature. But when they have a precise question, they often have to deal with too many documents to efficiently find the appropriate answers in a reasonable time. Faced to this literature overload, the need for automatic assistance has been largely pointed out, and PubMed is argued to be only the beginning on how scientists use the biomedical literature (Hunter and Cohen, 2006). Ontology-based search engines began to introduce semantics in search results. These systems still display documents, but the user visualizes clusters of PubMed results according to concepts which were extracted from the abstracts. GoPubMed (Doms and Schroeder, 2005) and EBIMed (Rebholz-Schuhmann et al, 2007) are popular examples of such ontology-based search engines in the biomedical domain. Question Answering (QA) systems are argued to be the next generation of semantic search engines (Wren, 2011). QA systems no more display documents but directly concepts which were extracted from the search results; these concepts are supposed to answer the user’s question formulated in natural language. EAGLi (Gobeill et al, 2009), our locally developed system, is an example of such QA search engines. Thus, both ontology-based and QA search engines, share the crucial task of efficiently extracting concepts from the result set, i.e. a set of documents. This task is sometimes called macro reading, in contrast with micro reading – or classification, categorization – which is a traditional Natural Language Processing task that aims at extracting concepts from a single document (Mitchell et al, 2009). This paper focuses on macro reading of MEDLINE abstracts. Several experiments have been reported to find the best way to extract ontology terms out of a single MEDLINE abstract, i.e. micro reading. In particular, (Trieschnigg et al, 2009) compared the performances of six classification systems for reproducing the manual Medical Subject Headings (MeSH) annotation of a MEDLINE abstract. The evaluated systems included two morphosyntactic classifiers (sometimes also called thesaurus-based), which aim at literally finding ontology terms in the abstract by alignment of words, and a machine learning (or supervised) classifier, which aims at inferring the annotation from a knowledge base containing already annotated abstracts. The authors concluded that the machine learning approach outperformed the morphosyntactic ones. But the macro reading task is fundamentally different, as we look for the best way to extract then combine ontology terms from a set of MEDLINE abstracts. The issue investigated in this paper is: to what extent the differences observed between two classifiers for a micro reading task are observed for a macro reading one? In particular, the redundancy hypothesis claims that the redundancy in large textual collections such as the Web or MEDLINE tends to smoothe performance differences across classifiers (Lin, 2007). To address this question, we compared a morphosyntactic and a machine learning classifiers for both tasks, focusing on the extraction of Gene Ontology (GO) terms, a controlled vocabulary for the characterization of proteins functions. The micro reading task consisted in extracting GO terms from a single MEDLINE abstract, as in the Trieschnigg et al’s work; the macro reading task consisted in extracting GO terms from a set of MEDLINE abstracts in order to answer to proteomics questions asked to the EAGLi QA system.

Background This paper describes and evaluates a sentence selection engine that extracts a GeneRiF (Gene Reference into Functions) as defined in ENTREZ-Gene based on a MEDLINE record. Inputs for this task include both a gene and a pointer to a MEDLINE reference. In the suggested approach we merge two independent sentence extraction strategies. The first proposed strategy (LASt) uses argumentative features, inspired by discourse-analysis models. The second extraction scheme (GOEx) uses an automatic text categorizer to estimate the density of Gene Ontology categories in every sentence; thus providing a full ranking of all possible candidate GeneRiFs. A combination of the two approaches is proposed, which also aims at reducing the size of the selected segment by filtering out non-content bearing rhetorical phrases. Results Based on the TREC-2003 Genomics collection for GeneRiF identification, the LASt extraction strategy is already competitive (52.78%). When used in a combined approach, the extraction task clearly shows improvement, achieving a Dice score of over 57% (+10%). Conclusions Argumentative representation levels and conceptual density estimation using Gene Ontology contents appear complementary for functional annotation in proteomics.