Developing an Integrated Rare Disease Bioinformatics Resource to Determine Phenotype to Genotype Correlations

该团队手动策划了两种罕见疾病，肌酸转运蛋白缺乏症（CTD）和法伯病（FD），以前曾在TDB临床前开发项目组合中工作。测试数据集用于验证和训练基于自然语言处理（NLP）的人工智能（AI）算法，以搜索PubMed中的罕见疾病基因变异关联。在此期间，该团队发布了Web应用程序的1.0版本。确定了125个以上的公共数据源，并不断将其纳入数据平台。罕见疾病的命名和本体已与遗传和罕见疾病信息中心（GARD）数据库协调一致，以便对感兴趣的基因、变异和疾病进行统一检索。该团队还实施了分析和可视化工具，以允许对2D/3D蛋白质结构和基因变体进行详细的可视化和交互式探索。此外，对CTD/SLC 6A 8和FD/ASAH 1基因数据的所有已发表、手动整理的文献进行了全面整合，以提供高准确度的基因型-表型相关性数据。 文献人工智能功能的开发最初的目标是挖掘相关的罕见疾病细节，并使研究人员、临床医生和患者护理人员能够访问这些信息。目前的用户界面允许挖掘已发表的同行评审文献的摘要，并查询基因和/或疾病及其关联。然后，人工智能算法检索与这些输入相关的出版物，更深入的探索可以揭示遗传变异和靶蛋白结构的分子细节。新型自然语言处理模型的定制实现将有助于评估模型对罕见疾病研究的适用性。还开始了纳入罕见病动物模型模块的工作，并提出了在罕见病研究中注入公平性的领域。目前正在努力建立一个全面的标准操作程序和方法，用于访问和纳入来自非公共数据库和来源的基因组变异和临床数据。

The team manually curated two rare diseases, Creatine Transporter Deficiency (CTD) and Farber Disease (FD), previously worked on in the TDB preclinical development project portfolio. The test data sets were used to validate and train natural language processing (NLP)-based artificial intelligence (AI) algorithms to search PubMed for rare disease gene variant associations. During this period, the team released version 1.0 of the web application. Over 125 public data sources were identified and are continuously being integrated into the data platform. The nomenclature and ontologies for the rare diseases have been harmonized with the Genetic and Rare Diseases Information Center (GARD) database to allow for harmonized searches for gene, variant and disease of interest. The team also implemented analytical and visualization tools to allow detailed visuals and interactive exploration of 2D/3D protein structures and gene variants. Additionally, all published, manually curated literature on CTD/SLC6A8 and FD/ASAH1 gene data was fully integrated to provide high accuracy genotype-phenotype correlation data. The development of the literature AI feature was initiated with the goal of mining relevant rare disease details and making the information accessible to researchers, clinicians, and patient caregivers. The current user interface allows mining of abstracts of published, peer-reviewed literature and query of a gene and/or disease and their association. AI algorithms then retrieve publications associated with these inputs, and deeper exploration can reveal molecular details of genetic variation and target protein structure. Customized implementation of novel natural language processing models will help assess the fit of the models to rare disease research. Work also has been initiated to incorporate rare disease animal model modules and present areas for infusing equity in rare disease research. Efforts are ongoing to establishing a comprehensive standard operating procedure and methodology for accessing and incorporating genomic variant and clinical data from non-public databases and sources.