A Pipeline to Automate the Updating of a Specialized Protein Database

Motivation: The growing number of specialized databases in molecular biology, coupled with the huge increase in the availability of molecular data, necessitates the development of automatic methods for finding and adding relevant information to these databases. Results: We show how a general protein database (Swiss-Prot) can be used as a source of data for a more specialized one (TCDB, the Transport Classification Database). First, we present a maximumentropy classification method trained on preprocessed Swiss-Prot records that achieves high precision and recall in determining which records are relevant to transmembrane transport in cross-validation experiments. Next, we describe a set of rules that can be used to further filter out proteins that are not novel, or not well characterized. Using both these pipeline stages, a human expert only has to examine about 2% of Swiss-Prot records for potential inclusion in