MACE2K - Molecular And Clinical Extraction: A Natural Language Processing Tool for Personalized Medicine

 描述(由申请人提供)：来自相关信息源的数据的速度、种类、数量和准确性使肿瘤学家收集和审查能够支持患者常规个性化治疗的相关数据具有极大的挑战性。迫切需要开发数据辩论方法，包括自然语言处理和信息检索方法，以摘录和整理与个性化治疗相关的出版物和临床试验。一旦经过整理，这些结构化的数据就可以被生物医学研究人员用来生成新的科学假设，设计新的研究，更好地了解疾病的生物学机制，执行荟萃分析，并创建临床决策支持系统。迫切需要开发特定于个性化治疗领域的改进的搜索界面，包括通过以下方式显示、排序和保存结果 最终用户。虽然存在几种基于数据库和网络的关键字搜索引擎算法，但缺乏满足个性化医疗的独特挑战的工具。还迫切需要开发软件，允许验证和确认通过使用主题专门知识的计算方法提取和排序的信息，以改进可用于个性化治疗的生物医学研究的黄金标准语料库。为了解决这些问题，我们将构建一个创新的软件堆栈(MACE2K)，以适应和扩展广泛测试的生物创新自然语言处理(NLP)工具，以自动检索和预处理来自临床试验网站、PubMed摘要以及开放获取文章和会议记录的靶向治疗信息。我们将建立一个实体提取盒来准确解析基因突变、易位、基因表达、蛋白质表达和蛋白质磷酸化。将建立一个标记消歧盒，以评估试验纳入或排除标准，并确定与标记有关的主要终点。我们将包括一个排序盒，它使用关于标志物、药物和试验的消除歧义的信息，根据试验和研究与个性化药物的相关性提供严格的评分。将建立一个新的游戏化墨盒，以使主题专家能够核实和验证信息语料库。我们的研究利用国家癌症研究所在几个项目上的投资(我们参与了其中许多项目)，包括NCI药物词典、国家癌症信息学计划(NCIP)、I-SPY试验和癌症系统生物学中心(CCSB)，以有效地实现我们的目标。

 DESCRIPTION (provided by applicant): The velocity, variety, volume and veracity of data from relevant information sources make it extremely challenging for oncologists to collect and review pertinent data that can support routine personalized treatment for their patients. There is an urgent need to develop data wrangling approaches including Natural Language Processing and information retrieval methods to extract and curate personalized-therapy related publications and clinical trials. Once curated, the structured data can be used by biomedical researchers to generate novel scientific hypotheses, design new studies, obtain a better understanding of biological mechanisms of disease, perform meta-analyses, and create clinical decision support systems. There is an urgent need to develop improved search interfaces specific to the field of personalized therapy, including ways to display, rank, and save results by end users. While several database and web-based keyword search engine algorithms exist, there is a lack of tools that meet the unique challenges of personalized medicine. There is also an urgent need to develop software that allows for verification and validation of information extracted and ranked through computational methods using subject matter expertise to improve the gold standard corpus that can be used for biomedical research into personalized therapies. To address these issues, we will build an innovative software stack (MACE2K) to adapt and extend widely tested Biocreative natural language processing (NLP) tools to automatically retrieve and pre-process targeted therapy information from clinicaltrials.gov, PubMed abstracts as well as open access articles, and conference proceedings. We will build an entity extraction cartridge to accurately parse gene mutations, translocations, gene expression, protein expression, and protein phosphorylation. A marker disambiguation cartridge will be built to assess for trial inclusion or exclusion criteria and to determine marker-related primary endpoints. We will include a ranking cartridge that uses the disambiguated information on markers, drugs and trials to provide a rigorous scoring of trials and studies according to their relevance for personalized medicine. A novel gamification cartridge will be built to allow subject matter experts to verify and validate the information corpus. Our research leverages National Cancer Institute's investments in several programs (many of which we are involved in) including the NCI drug dictionary, National Cancer Informatics Program (NCIP), I-SPY trials, and Center for cancer systems biology (CCSB) to efficiently accomplish our aims.