Assigning Roles to Protein Mentions: the Case of Transcription Factors Export

@article{citeulike:4386450, abstract = {Transcription factors (TFs) play a crucial role in gene regulation, and providing structured and curated information about them is important for genome biology. Manual curation of TF related data is time-consuming and always lags behind the actual knowledge available in the biomedical literature. Here we present a machine-learning text mining approach for identification and tagging of protein mentions that play a TF role in a given context to support the curation process. More precisely, the method explicitly identifies those protein mentions in text that refer to their potential TF functions. The prediction features are engineered from the results of shallow parsing and domain-specific processing (recognition of relevant appearing in phrases) and a phrase-based Conditional Random Fields (CRF) model is used to capture the content and context information of candidate entities. The proposed approach for the identification of TF mentions has been tested on a set of evidence sentences from the TRANSFAC and FlyTF databases. It achieved an F-measure of around 51.5\% with a precision of 62.5\% using 5-fold cross-validation evaluation. The experimental results suggest that the phrase-based CRF model benefits from the flexibility to use correlated domain-specific features that describe the dependencies between TFs and other entities. To the best of our knowledge, this work is one of the first attempts to apply text-mining techniques to the task of assigning semantic roles to protein mentions.}, author = {Yang, Hui and Keane, John and Bergman, Casey M. M. and Nenadic, Goran}, citeulike-article-id = {4386450}, citeulike-linkout-0 = {http://dx.doi.org/10.1016/j.jbi.2009.04.001}, citeulike-linkout-1 = {http://linkinghub.elsevier.com/retrieve/pii/S1532046409000525}, citeulike-linkout-2 = {http://view.ncbi.nlm.nih.gov/pubmed/19364541}, citeulike-linkout-3 = {http://www.hubmed.org/display.cgi?uids=19364541}, day = {10}, doi = {10.1016/j.jbi.2009.04.001}, issn = {15320464}, journal = {Journal of Biomedical Informatics}, keywords = {abner, biomedical, context-features, crf, crfplusplus, curation, dependencies, genia-tagger, information-extraction, keywords, lingpipe, local-features, long-distance, machine-learning, named-entities, parsing, pattern-learning, phrasal-representation, phrase-window, phrases, semantic-features, semantic-roles, sentence-classification, shallow-parsing, syntactic-features, transcription-factor}, month = {October}, number = {5}, pages = {887--894}, posted-at = {2009-04-23 14:46:51}, priority = {0}, title = {Assigning Roles to Protein Mentions: the Case of Transcription Factors}, url = {http://dx.doi.org/10.1016/j.jbi.2009.04.001}, volume = {42}, year = {2009} }

TY - JOUR ID - citeulike:4386450 L3 - citeulike-article-id:4386450 N2 - Transcription factors (TFs) play a crucial role in gene regulation, and providing structured and curated information about them is important for genome biology. Manual curation of TF related data is time-consuming and always lags behind the actual knowledge available in the biomedical literature. Here we present a machine-learning text mining approach for identification and tagging of protein mentions that play a TF role in a given context to support the curation process. More precisely, the method explicitly identifies those protein mentions in text that refer to their potential TF functions. The prediction features are engineered from the results of shallow parsing and domain-specific processing (recognition of relevant appearing in phrases) and a phrase-based Conditional Random Fields (CRF) model is used to capture the content and context information of candidate entities. The proposed approach for the identification of TF mentions has been tested on a set of evidence sentences from the TRANSFAC and FlyTF databases. It achieved an F-measure of around 51.5% with a precision of 62.5% using 5-fold cross-validation evaluation. The experimental results suggest that the phrase-based CRF model benefits from the flexibility to use correlated domain-specific features that describe the dependencies between TFs and other entities. To the best of our knowledge, this work is one of the first attempts to apply text-mining techniques to the task of assigning semantic roles to protein mentions. IS - 5 JF - Journal of Biomedical Informatics SN - 15320464 EP - 894 TI - Assigning Roles to Protein Mentions: the Case of Transcription Factors VL - 42 SP - 887 KW - abner KW - biomedical KW - context-features KW - crf KW - crfplusplus KW - curation KW - dependencies KW - genia-tagger KW - information-extraction KW - keywords KW - lingpipe KW - local-features KW - long-distance KW - machine-learning KW - named-entities KW - parsing KW - pattern-learning KW - phrasal-representation KW - phrase-window KW - phrases KW - semantic-features KW - semantic-roles KW - sentence-classification KW - shallow-parsing KW - syntactic-features KW - transcription-factor AU - Yang, Hui AU - Keane, John AU - Bergman, Casey AU - Nenadic, Goran PY - 2009/10/10/ UR - http://dx.doi.org/10.1016/j.jbi.2009.04.001 ER -