Information extraction from semi-structured data in the protein data bank by induction of a data description pattern

PDB (Protein Data Bank) is a primary database that stores the three-dimensional data of a protein structure. This paper proposes a system, the PDB REMARK transcoder, that semi-automatically extracts significant data from REMARK lines, a part of the PDB data, and transcodes them to XML (eXtensible Markup Language) format. This system induces a description pattern from some protein entries to accept gradual variations of REMARK lines. Tokens (words and phrases) are clustered by evaluating their similarity using token attributes, and their contents are recognized by cluster labels. By using finite state automatons, description patterns are induced, and then iterative structures are correspondly nested into XML formats. The confidence of the output XML data is confirmed by log files. Applying the system to the REMARK lines of 8,906 protein entries clarified the effectiveness of the method.