The core technologies we are developing are based on ontologies and the Semantic Web. In information technology, an ontology semantically defines domain knowledge in standard ways that allow computer programs to understand and query the knowledge intelligently. Ontologies are commonly used in the biomedical domain. Heterogeneous data can be naturally linked together through domain ontologies. An ontology-based knowledgebase allows intelligent and semantic queries supported by the Semantic Web technologies. Dr. Tao has extensive experiences in ontology generation and conceptual modeling. She has proposed a set of terminology representation guidelines for biomedical ontologies to increase semantic interoperability. The relevant technologies have been applied to large federally funded projects such as SHARPc, NCI terminology services, NCBO, and BioCADDIE. Dr. Cui has strong grounding in algorithms and computational methodologies for ontology quality control. For instance, she has designed and developed scalable algorithms for analyzing biomedical ontologies, including computing non-lattice subgraphs and detecting relation reversals in the SNOMED CT, mining lexical patterns in non-lattice subgraphs for detecting missing hierarchical relations and concepts in SNOMED CT, Gene Ontology and NCI Thesaurus, performing cross-ontology hierarchical relation examination in the Unified Medical Language System, and mining diverse clinical datasets from the National Sleep Research Resource.
Ontologies have been widely used in different biomedical applications. Biomedical ontologies can be designed for different purposes such as knowledge representation for a specific domain, formal classification for domain concepts, standardized and normalized representation for domain concepts and relations, and semantic reasoning for different applications such as clinical decision support, drug repurposing, risk prediction, and phenotyping. Our group has been working on different projects focusing on biomedical ontology applications. For example, we have developed an approach for ontology-based drug target inference and has applied this approach to colorectal cancer and Alzheimer’s disease. We have been developing ontology-based health dialogue systems to improve clinician and consumer communications. We propose to construct a set of (1) domain ontologies to equip the virtual agent with knowledge surrounding specific health-related topics and (2) application ontologies that regulate the behavior of the conversational agent. In addition, we have also been working on developing a time ontology (Time Event Ontology) to formally represent temporal information and relations to enable semantic reference.
As we are entering the big data era, a key challenge in biomedical research is the complexity and heterogeneous nature of its data. Ontology-based technologies can provide an effective way to harmonize the data for further analyses. We have been focusing on developing novel machine learning and statistical methods to handle different types of data from heterogeneous sources to solve healthcare, biomedical, or public health problems. For example, using EHR data, we are developing risk prediction algorithms for different clinical conditions such as asthma exacerbation and myocardial infarction. We are also working toward precision medicine applications such as to develop novel deep learning algorithms for personalized recommendation for antiplatelet therapy duration for patients underwent coronary stent implantation. Using data from the health departments, we are trying to develop a novel deep learning algorithm to predict HIV infection risks for the young MSM population. Using FDA’s postmarket surveillance data, we are developing statistical methods for drug and vaccine safety analyses. In addition, we also have extensive experiences on analyzing social media data for public perception studies and personalized intervention.