Seq2Ref: a web server to facilitate functional interpretation

Wenlin Li, Qian Cong, Lisa N. Kinch, Nick V. GrishinBMC Bioinformatics 2013 · DOI

TL;DR

Seq2Ref is a new web server that helps researchers interpret the function of unknown proteins by identifying and ranking experimentally characterized "reference proteins" that are homologous to the query. It leverages BLAST, literature mining, and a novel scoring system to provide a more reliable basis for functional annotation than current automated methods.

What Problem Does This Paper Address?

The exponential growth of protein sequence data from genome sequencing projects has outpaced the ability to experimentally characterize each protein. Consequently, the vast majority of protein annotations are derived from computational predictions, which are prone to errors and can be outdated. This makes it difficult for researchers to gain a deep understanding of protein function, especially for novel or hypothetical proteins. Existing annotation databases often provide insufficient detail, and while cross-linking proteins to literature is possible, identifying truly relevant "reference proteins" (those with experimental validation) from the sheer volume of data is a significant challenge.

Methods

Seq2Ref employs a multi-step approach:

  1. Homology Detection: It uses BLAST to find proteins homologous to the query sequence in the NCBI nr database.
  2. Reference Protein Identification: From the BLAST hits, it identifies "reference proteins" by looking for those linked to PubMed literature, present in the Protein Data Bank (PDB), or in Swiss-Prot (SP). These databases are prioritized because they contain experimentally validated or manually curated information.
  3. Homology Evaluation and Ranking: A novel plurality-based rating system is used to evaluate the relevance of these reference proteins. This system integrates:
  4. Reciprocal BLAST: Checks if the identified homolog also finds the query as its best hit, indicating a strong evolutionary relationship.
  5. Multiple Sequence Comparison (MSC): Compares the query to a set of closely related sequences to gain more robust homology statistics than simple pairwise comparisons.
  6. Pairwise Comparison: Assesses the similarity between the query and the reference protein.
  7. Reference Protein Status: A bonus point is given if the hit is already designated as a reference protein (e.g., has PubMed links).

These features are combined to assign a score, with higher scores indicating a greater likelihood of functional similarity.

Datasets

  • NCBI nr database: Used for initial homology searches.
  • Protein Data Bank (PDB): Source of experimentally determined protein structures.
  • Swiss-Prot (SP): A curated protein sequence database.
  • PubMed: Used to retrieve publications associated with reference proteins.
  • Enolase superfamily: Used as a benchmark dataset to evaluate the rating system's accuracy and coverage.

Key Insights and Ideas

  • The critical need for "reference proteins": The paper strongly advocates for relying on experimentally validated proteins for functional interpretation, highlighting the unreliability of purely computational annotations.
  • A novel scoring system for relevance: The plurality-based rating system, integrating reciprocal BLAST, MSC, and other features, is a key innovation. It moves beyond simple BLAST scores to provide a more nuanced assessment of functional relevance.
  • Bridging the gap between sequence and function: Seq2Ref aims to make the wealth of experimental data and literature accessible to researchers working with uncharacterized proteins, facilitating hypothesis generation.
  • The challenge of identifying relevant literature: Even with cross-links, finding the most relevant publications for a specific query protein is non-trivial, and Seq2Ref attempts to automate this.

Key Results

  • The benchmark using the enolase superfamily showed that the rating system achieves 100% accuracy at a score cutoff of 4 for family-level interpretation and 100% coverage for subgroup-level interpretation.
  • The server successfully identified relevant reference proteins in case studies, providing more detailed functional insights than standard database annotations. For example, it helped link a hypothetical protein to potential tumor suppressor activity and identified a likely role for a protein in membrane insertion.
  • Seq2Ref provides a ranked list of reference proteins, allowing users to prioritize their investigation.

Limitations and Future Work

  • Quality of NCBI cross-links: The accuracy of Seq2Ref is dependent on the quality and correctness of the links between NCBI Protein and PubMed databases. Errors or missing links could lead to missed relevant literature or incorrect assignments.
  • Direct transfer of function: While Seq2Ref aims to facilitate functional interpretation, the authors caution that direct transfer of specific functions (like substrate specificity) from a hit to a query should be done carefully and requires further verification.
  • Orthology verification: Verifying true orthology often requires additional, diligent analysis beyond the scope of the server.
  • Future work: The authors suggest that critical inspection of publications and homology is essential for the best conclusions about protein function. The server's utility in providing a basis for experimental design and understanding novel proteins is highlighted.

Summary updated

BibTeX
@article{li2013seq,
  author = {Wenlin Li and Qian Cong and Lisa N. Kinch and Nick V. Grishin},
  title = {{Seq2Ref: a web server to facilitate functional interpretation}},
  journal = {BMC Bioinformatics},
  year = {2013},
  doi = {10.1186/1471-2105-14-30},
}
Figure