We establish a network of excellence on biomedical information management among some of the Danube States, and we strive to extend the initiative to multiple European countries.
Within current AI research, in BIMDANUBE, we aim to center the human within a machine learning setting, which enables interactivity between the human and the machine. The Human-In-The-Loop paradigm allows a user to take actions in order to influence a machine learning model, which improves over time through usage.
Targeting the health domain, we work closely together with experts and medical doctors and strive to a better understanding of the vast amount of data available.
During the course of BIMDANUBE, we are going to organize two workshops on current and future information management in the biomedical domain, we conduct a feasibility study on a novel bottom-up approach to information management. Our continuously growing network will prepare an EU grant proposal on novel methodologies in biomedical information management for practicioners and teaching purposes.
Recently, knowledge management as a field faced several challenges. On one hand, sophisticated technologies and standards were developed to support knowledge-based modeling, such as domain ontologies including MeSH, Disease Ontology, and Gene Ontology and the Sematic Web description languages and infrastructures including RDF, OWL, SPARQL and others. On the other hand, however the current approaches face three major issues: (1) knowledge bottleneck: required resources for knowledge management such as domain ontologies are not available for many domains and languages; (2) the overall approach of knowledge management did not get widely spread due to the fact that it imposes a large burden on the user, such as annotation or expertise with complex tools like Protégé; (3) modeling entire domains as large as the medical domain with (English-oriented) knowledge resources does not meet requirements of users, who are mostly specializing in a certain sub-field and also need to operate in their local language.
We are reloading this traditional heavyweight top-down knowledge management approach and replace it with a much simpler and practical problem-oriented bottom-up approach. We choose the biomedical domain as a playground for our experiments as this domain concerns humankind as a whole, thus being of high priority with regard to the EU Horizon 2020 strategic plan. Further, we deem the Danube region with its high variability in languages and heterogeneity as an ideal test bed for our approach.
Medical researchers have to process an enormous amount of the literature – PubMed adds about half a million paper to its index each year. Literature search and reasoning is demanding, because of the need to revealing and maintaining many complex relationships between numerous sets of entities. In order to alleviate the efforts of biomedical research related to literature we propose a novel conception to information management based on bottom-up construction of a problem-oriented ontology, called entity graph (EG). Entity graphs provide a new tool for medical researchers that (1) help to document relations between biomedical entities in a compact intuitive and interpretable form; (2) generate new relations in a semi-automatic way based on corpus analysis; (3) communicate new biomedical knowledge in a form of an easily interpretable interactive graph and (4) share knowledge and annotations amongst researchers.
The workshop will gather interested professionals from the Danube region and across Europe working in the information or medical domain, such as medical researchers, medical doctors and entrepreneurs building their business around biomedical ICT. The goal of the workshop is to obtain answers to the following questions:
Besides the invited speakers, the workshop will feature regular participants from the Danube states who are interested in the subject of biomedical information management and keen to establish a new project in this area.
|WHEN||19. & 20. Feb. 2018|
|WHERE||University of Hamburg, Hamburg, Germany|
More information is availbable on the official workshop hompage.
The event will be more technical in nature than the first workshop. The goal of the second workshop will be to answer on the following questions:
|WHEN||TBD (~Jul. 2018)|
The team at the University of Hamburg with help of collaborators will drive this study. The goal of the experiment is to try to reload the traditional top-down knowledge management approaches, involving large domain ontologies. The entity graph can be considered as a personalized problem-oriented ontology, created during investigation of a narrow problem. The main advantage of the entity graph is that it is much simpler, easy-to-use and goal-oriented with respect to the classical ontologies and other heavyweight knowledge management techniques. During the second workshop, we are going to present and discuss bottom-up approach to biomedical information management devised in this study.
The team at the University of Hamburg will build upon multiple prior developments to perform this study efficiently. This team is going to implement the software prototype required for the experiment and for the overall design of the experiment. The feasibility study will be conducted on biomedical researchers from the partner institutions in Austria, Croatia and Bulgaria.
Our prototype presents a novel way to deal with biomedical knowledge based on entity graphs, thus introducing and evaluating a novel human-computer interaction paradigm for personal and shared information management. This new paradigm, in combination with the free availability of our prototype, might have a disruptive effect on the content management and knowledge/information management industries, which still largely operate in ontology/taxonomy/vocabulary-driven ways. The new concept of bottom-up information management can create an ecosystem of providers that offer software solutions for the integration of this concept with existing systems. Further, there are considerable business opportunities in maintaining distributed repositories, as well as interlinking them.