Datascope: Mining Biological Sequences

MUCH OF THE WORK IN THE data-mining community defines mining data as a collection of techniques for extracting knowledge out of large databases. This definition is a bit ambiguous because the “knowledge” extracted from databases varies dramatically across systems. In the spirit of intelligent systems, I suggest we define data mining roughly as a collection of techniques that produce representations that efficiently support a class of queries about the original database. I argue that the ability to efficiently and accurately answer a wide range of queries is a plausible measure of “understanding” the data. By queries, I do not mean traditional queries. The exact definition of the query language depends on the application. Our informal definition suggests the need to develop formal criteria that measure an intelligent system’s effectiveness in terms of its ability to efficiently and accurately construct local representations, integrate these representations with a global architecture, and answer complex questions about the domain. Mining biological databases is a particularly important challenge for this philosophy. The ultimate goal of mining biological sequences is to develop a comprehensive computational framework that can take a genomic DNA sequence and automatically produce a detailed annotation of the organism—one that describes all the genes, how the genes are regulated by regulatory elements, and how the genes function together to produce higher function and behavior levels. The research in the intelligent bioinformatics community focuses on learning algorithms, data-mining tools, and systems that will let us transform biological sequences, observations, and knowledge into structured and meaningful information that scientists can query, visualize, and understand. I call this approach to mining biological sequences Datascope, because it focuses on providing biologists with the capability of probing genomic data with various degrees of detail and a wide range of viewpoints. The approach produces systems that help discover the fundamental connections between genetic sequences and functions of living organisms. Computational-genomics problems