Links and Paths through Life Sciences Data Sources

An abundance of biological data sources contain data on classes of scientific entities, such as genes and sequences. Logical relationships between scientific objects are implemented as URLs and foreign IDs. Query processing typically involves traversing links and paths (concatenation of links) through these sources. We model the data objects in these sources and the links between objects as an object graph. We identify a set of interesting properties for links and paths, such as outdegree, image of a link, cardinality of data objects and links, the number of distinct objects reached by some links, etc. Analogous to database cost models, we use statistics from the object graph to develop a framework to estimate the result size for a query on the object graph. Analogous to training and testing, we use sampled data from queries to estimate the result size. We validate our models using data sampled from four NIH/NCBI data sources. Our research provides a foundation for querying and exploring data sources. 1 Querying Interlinked Sources An abundance of biological data sources contain data about scientific entities, such as genes and sequences. Each source may have data on one or more logical classes. Logical relationships between scientific objects are implemented as source links between data sources. Together, they form a graph – the source graph. Each source link represents a collection of object links, each going from a data object in one source to another object, in the same or a different source. An object graph is formed of the data objects and links. Formal definitions are in Sec. 2. Scientists are interested in exploring relationships between scientific objects, e.g., genes and citations. Consider the query “Return all citations of PubMed that are linked to an Omim entry