The Identification of Target Proteins from Patents - Mining of biological entities from a full-text patent database
Year: 2010
Type: dissertation
Abstract: Modern drug discovery involves identifying therapeutic targets (i.e.enzymes, receptors, and other proteins) that are relevant to a disease of interest, and searching for small-molecule compounds that are able to regulate these targets in order to cure or relieve the disease.Therefore, target-assay-compound relationships are key technical information driving modern drug discovery processes.As a consequence, every pharmaceutical R&D organization must seek for this information both from internal and external sources.One of the important information sources is patent documents which were disclosed for public access largely by pharmaceutical enterprises and academic institutions.While the target-assay-compound relationships have long been extracted from patents manually by expert curators, progress in text-mining shows possibilities to automate the extraction.Although the accuracy of human experts cannot yet be matched by automated text mining, access to a substantial corpus of full-text patents is clearly of complementary value for its scale, coverage, and speed.This work presents results from investigating a full-text patent resource, that AstraZeneca has access to via a collaboration consortium arrangement with IBM.The work focused on exploring challenges and possible solutions for retrieval of target-containing patents and extraction of target protein names from individual patents.The work also serves as a pilot project in exploring and pioneering the mining of structure activity relationship (SAR) data from patents.Several information retrieval and extraction approaches were benchmarked against a source of SAR information manually extracted from patents which is a licensed product from GVKBIO.It was shown that targetcontaining patents could be retrieved by using keywords in titles with acceptable recall and precision.It was also shown that proximal co-occurrence between protein names and chemical modulation keywords could be used to identify target protein names in full-text patents with different recall and precision particular to patent sections.Assessments of false positives and false negatives from these approaches suggested that the extraction of target protein names required advance text mining techniques in order to interpret semantic context.Furthermore, it was also suggested that the writing practice of pharmaceutical patents is less scientific compared to biomedical journals.This leads to several issues in information retrieval and extraction specific to pharmaceutical patents.In summary, this work has successfully explored possible solutions for the retrieval of target-containing patents and the extraction of target protein names, and paints the way forward for SAR data mining from pharmaceutical patents.
Author: Ithipol Suriyawongkul
Language: English
Subfield: Health Informatics
Field: Medicine
Domain: Health Sciences
Open Access status: green
Funder: AstraZeneca
OpenAlex ID: W2117727240