In proteomics workflows, proteins are often digested first, then peptides are separated and subjected to identification by mass spectrometry (e.g., 2D-LC). In this process the peptide assignment to a protein is lost and has to be rebuilt by bioinformatic methods. We present ProteinExtractor, a module of the ProteinScape Bioinformatics Platform, which uses an empiric, iterative method to derive minimal protein lists from peptide search results, which may even come from different search algorithms or different MS datasets. ProteinExtractor uses an iterative approach to generate a minimal protein list. With composite database searches ProteinExtractor allows measuring the false-positive rate of the protein list. A test dataset (five recombinant proteins, 408 spectra, Bruker Ultraflex), and a real-life dataset (200410 LC/ESI-MS/MS spectra, Bruker Esquire HCT-Ultra, and 11619 LC/MALDI-MS/MS spectra, Bruker Ultraflex, both obtained from an analysis of proteins from a human cell line—SW480) were analyzed. The most probable protein sequence entries contained in the test dataset were identified with intensive manual data interpretation by several mass spectrometry experts. Using standard search algorithms, the correct protein sequence database entries are scattered over the first 171 protein ranks. Together with application specialists, we developed a set of rules to define a minimal protein list containing only those proteins (and isoforms) that can be unequivocally distinguished on the basis of MS/MS data. Applying these rules, the correct five proteins are ranked within the top eight protein candidates. In the real-life dataset, the peptide search results of Mascot, Sequest, Phenyx, and ProteinSolver were merged using ProteinExtractor. Merging all four search algorithms, over 50% more proteins could be identified than by using Mascot alone (with a false-positive rate of less then 2.5%). Merging ESI and MALDI data together, another 25% more proteins could be identified.

The data acquisition phase of initial pilot studies (human and mouse brain samples) of the Human Proteome Organisation (HUPO) Brain Proteome Project (BPP) is now complete and the data generated by the participating laboratories has been submitted to the central Data Collection Center. The BPP Bioinformatics Group met on 8th April 2005 at the European Bioinformatics Institute (Hinxton, UK) to discuss strategies for the reanalysis of the pooled data from all the participating laboratories. A summary of the results of the data reprocessing will be presented at the 4th HUPO World Congress that will be held in August/September 2005.