PRECIS: an automated pipeline for producing concise reports about proteins

There have been several attempts at addressing the problem of annotating sequence data computationally. Annotation generation can be considered a pipeline of processes: first harvesting data from a variety of data sources, then distilling and transforming it into a form more appropriate for the end database. This task is usually performed by human annotators, a solution that is clearly not scaleable. There have been several attempts to mimic some of these pipelines in software. However, these have generally focused on low level annotation, such as database cross-references, or by harvesting data from computational techniques such as gene finding or similarity searches. Higher level annotation such as that seen in the PRINTS database is usually formed from data that is free text, or only partly structured. This presents a much greater computational challenge. Therefore we studied the pipeline that is used to generate annotation for the PRINTS database, and have developed prototype software that reflects and automates this pipeline. As this software operates primarily on data culled from the SWISS-PROT database, we have called it PRECIS (Protein Reports Engineered from Concise Information in SWISS-PROT). This software is currently being used to generate annotation for the prePRINTS database. As the output is a structured report detailing the function, structure and disease associations of a protein, and providing literature references and keywords we believe it will be of more generic use. The software is available on request from mitchell@bioinf.man.ac.uk.