Solutions

Fusion for gene prioritization

One of the grand challenges in study design and interpretation is the fusion of heterogeneous information sources. In a practical approach the goal and the context of the fusion is simplified to generate an ordering of the entities, e.g. genes, reflecting their consensual relevance to a given problem according to the information sources.

Two radically different approaches formed for the generation of integrated ranking. The numeric approaches integrate the data (data fusion), integrate statistics of the data (e.g. kernel fusion), or they integrate the results of the analysis of the separate information sources (e.g. rank fusion). The range of semantic approaches starts from relational databases integration and ends at natural language query processing systems over knowledge bases.
The main shortcomings of numeric fusion methods are their inability to cope with contextuality, whereas the semantic/knowledge based approach cannot incorporate uncertain information and it depends on the experience of the user for formulating queries.

Our solutions based on probabilistic databases and probabilistic knowledge bases incorporating the results of statistical data analysis and background knowledge provide remedies for many such problems. These allow full control over contextualities and at the same time they offer complex query schemes over diverse background knowledge, such as free-text, controlled language, and logical knowledge elements.