Precision information extraction for rare disease epidemiology at scale.

BACKGROUND: The United Nations recently made a call to address the challenges of an estimated 300 million persons worldwide living with a rare disease through the collection, analysis, and dissemination of disaggregated data. Epidemiologic Information (EI) regarding prevalence and incidence data of rare diseases is sparse and current paradigms of identifying, extracting, and curating EI rely upon time-intensive, error-prone manual processes. With these limitations, a clear understanding of the variation in epidemiology and outcomes for rare disease patients is hampered. This challenges the public health of rare diseases patients through a lack of information necessary to prioritize research, policy decisions, therapeutic development, and health system allocations. METHODS: In this study, we developed a newly curated epidemiology corpus for Named Entity Recognition (NER), a deep learning framework, and a novel rare disease epidemiologic information pipeline named EpiPipeline4RD consisting of a web interface and Restful API. For the corpus creation, we programmatically gathered a representative sample of rare disease epidemiologic abstracts, utilized weakly-supervised machine learning techniques to label the dataset, and manually validated the labeled dataset. For the deep learning framework development, we fine-tuned our dataset and adapted the BioBERT model for NER. We measured the performance of our BioBERT model for epidemiology entity recognition quantitatively with precision, recall, and F1 and qualitatively through a comparison with Orphanet. We demonstrated the ability for our pipeline to gather, identify, and extract epidemiology information from rare disease abstracts through three case studies. RESULTS: We developed a deep learning model to extract EI with overall F1 scores of 0.817 and 0.878, evaluated at the entity-level and token-level respectively, and which achieved comparable qualitative results to Orphanet's collection paradigm. Additionally, case studies of the rare diseases Classic homocystinuria, GRACILE syndrome, Phenylketonuria demonstrated the adequate recall of abstracts with epidemiology information, high precision of epidemiology information extraction through our deep learning model, and the increased efficiency of EpiPipeline4RD compared to a manual curation paradigm. CONCLUSIONS: EpiPipeline4RD demonstrated high performance of EI extraction from rare disease literature to augment manual curation processes. This automated information curation paradigm will not only effectively empower development of the NIH Genetic and Rare Diseases Information Center (GARD), but also support the public health of the rare disease community.

Scientific evidence based rare disease research discovery with research funding data in knowledge graph.

BACKGROUND: Limited knowledge and unclear underlying biology of many rare diseases pose significant challenges to patients, clinicians, and scientists. To address these challenges, there is an urgent need to inspire and encourage scientists to propose and pursue innovative research studies that aim to uncover the genetic and molecular causes of more rare diseases and ultimately to identify effective therapeutic solutions. A clear understanding of current research efforts, knowledge/research gaps, and funding patterns as scientific evidence is crucial to systematically accelerate the pace of research discovery in rare diseases, which is an overarching goal of this study. METHODS: To semantically represent NIH funding data for rare diseases and advance its use of effectively promoting rare disease research, we identified NIH funded projects for rare diseases by mapping GARD diseases to the project based on project titles; subsequently we presented and managed those identified projects in a knowledge graph using Neo4j software, hosted at NCATS, based on a pre-defined data model that captures semantics among the data. With this developed knowledge graph, we were able to perform several case studies to demonstrate scientific evidence generation for supporting rare disease research discovery. RESULTS: Of 5001 rare diseases belonging to 32 distinct disease categories, we identified 1294 diseases that are mapped to 45,647 distinct, NIH-funded projects obtained from the NIH ExPORTER by implementing semantic annotation of project titles. To capture semantic relationships presenting amongst mapped research funding data, we defined a data model comprised of seven primary classes and corresponding object and data properties. A Neo4j knowledge graph based on this predefined data model has been developed, and we performed multiple case studies over this knowledge graph to demonstrate its use in directing and promoting rare disease research. CONCLUSION: We developed an integrative knowledge graph with rare disease funding data and demonstrated its use as a source from where we can effectively identify and generate scientific evidence to support rare disease research. With the success of this preliminary study, we plan to implement advanced computational approaches for analyzing more funding related data, e.g., project abstracts and PubMed article abstracts, and linking to other types of biomedical data to perform more sophisticated research gap analysis and identify opportunities for future research in rare diseases.

Phenotypically Similar Rare Disease Identification from an Integrative Knowledge Graph for Data Harmonization: Preliminary Study.

BACKGROUND: Although many efforts have been made to develop comprehensive disease resources that capture rare disease information for the purpose of clinical decision making and education, there is no standardized protocol for defining and harmonizing rare diseases across multiple resources. This introduces data redundancy and inconsistency that may ultimately increase confusion and difficulty for the wide use of these resources. To overcome such encumbrances, we report our preliminary study to identify phenotypical similarity among genetic and rare diseases (GARD) that are presenting similar clinical manifestations, and support further data harmonization. OBJECTIVE: To support rare disease data harmonization, we aim to systematically identify phenotypically similar GARD diseases from a disease-oriented integrative knowledge graph and determine their similarity types. METHODS: We identified phenotypically similar GARD diseases programmatically with 2 methods: (1) We measured disease similarity by comparing disease mappings between GARD and other rare disease resources, incorporating manual assessment; 2) we derived clinical manifestations presenting among sibling diseases from disease classifications and prioritized the identified similar diseases based on their phenotypes and genotypes. RESULTS: For disease similarity comparison, approximately 87% (341/392) identified, phenotypically similar disease pairs were validated; 80% (271/392) of these disease pairs were accurately identified as phenotypically similar based on similarity score. The evaluation result shows a high precision (94%) and a satisfactory quality (86% F measure). By deriving phenotypical similarity from Monarch Disease Ontology (MONDO) and Orphanet disease classification trees, we identified a total of 360 disease pairs with at least 1 shared clinical phenotype and gene, which were applied for prioritizing clinical relevance. A total of 662 phenotypically similar disease pairs were identified and will be applied for GARD data harmonization. CONCLUSIONS: We successfully identified phenotypically similar rare diseases among the GARD diseases via 2 approaches, disease mapping comparison and phenotypical similarity derivation from disease classification systems. The results will not only direct GARD data harmonization in expanding translational science research but will also accelerate data transparency and consistency across different disease resources and terminologies, helping to build a robust and up-to-date knowledge resource on rare diseases.