Published and Perished? The Influence of the Searched Protein Database on the Long-Term Storage of Proteomics Data

Abstract

In proteomics, protein identifications are reported and stored using an unstable reference system: protein identifiers. These proprietary identifiers are created individually by every protein database and can change or may even be deleted over time.To estimate the effect of the searched protein sequence database on the long-term storage of proteomics data we analyzed the changes of reported protein identifiers from all public experiments in the Proteomics Identifications (PRIDE) database by November 2010. To map the submitted protein identifier to a currently active entry, two distinct approaches were used. The first approach used the Protein Identifier Cross Referencing (PICR) service at the EBI, which maps protein identifiers based on 100% sequence identity. The second one (called logical mapping algorithm) accessed the source databases and retrieved the current status of the reported identifier.Our analysis showed the differences between the main protein databases (International Protein Index (IPI), UniProt Knowledgebase (UniProtKB), National Center for Biotechnological Information nr database (NCBI nr), and Ensembl) in respect to identifier stability. For example, whereas 20% of submitted IPI entries were deleted after two years, virtually all UniProtKB entries remained either active or replaced. Furthermore, the two mapping algorithms produced markedly different results. For example, the PICR service reported 10% more IPI entries deleted compared with the logical mapping algorithm. We found several cases where experiments contained more than 10% deleted identifiers already at the time of publication. We also assessed the proportion of peptide identifications in these data sets that still fitted the originally identified protein sequences. Finally, we performed the same overall analysis on all records from IPI, Ensembl, and UniProtKB: two releases per year were used, from 2005. This analysis showed for the first time the true effect of changing protein identifiers on proteomics data. Based on these findings, UniProtKB seems the best database for applications that rely on the long-term storage of proteomics data.

Crowdsourcing in proteomics: public resources lead to better experiments

“…The effect that this has to the proteomics community has been examined in some detail in two recent papers (Griss et al 2011a;Griss et al 2011b) where it was also shown that UniProt now provides directly comparable alternatives, an observation that is also evident for human sequences from Figure 3, where both UniProt and Ensembl are seen to closely resemble IPI.…”

Beware of Moving Targets: Reference Proteome Content Fluctuates Substantially Over the Years

“…As a result, alternative e®orts such as UniProtKB or the now discontinued International Protein Index (IPI) lead to distinctively di®erent results in the detail (di®erences with up to a¯fth of all entries) although, satisfyingly, the databases are largely overlapping. 11,12 This is not surprising since all large protein sequence databases are strongly interconnected via the International Nucleotide Sequence Database Consortium (INSDC) providing the EMBL/GenBank/DDBJ nucleotide set used by di®erent protein annotation pipelines. For example, about 98% of the protein sequences provided by UniProtKB come from the translations of coding sequences submitted to INSDC.…”

“…There is not only the issue of identi¯er stability over time. 11,12 This is a serious technical hindrance; yet, it can in principle be overcome by software adaptations as long as the sequence itself does still exist in the database. Similarly, the di®erences among alternative proteome databases for the same species can be circumvented by just applying all these proteome instances to the problem of study in parallel.…”

Exploring the potential of public proteomics data

“…This enabled the authors to look for the expression of specific splice isoforms from CNS‐related genes. Finally, in another example of PRIDE data reuse, UniProtKB was determined to be the most suitable reference database for long‐term proteomics data storage 46.…”

“…The gene and protein sequence databases that identification depends on are constantly evolving and improving 46. This means that reprocessing a proteomics dataset with an updated version of the gene or protein database can result in improved findings.…”