PGA: an R/Bioconductor package for identification of novel peptides using a customized database derived from RNA-Seq

Bo Wen, Shaohang Xu, Ruo Zhou, Bing Zhang, Xiaojing Wang, Xin Liu, Xun Xu, Siqi LiuBMC Bioinformatics 2016 · DOI

TL;DR

This paper introduces PGA, an R/Bioconductor package that automates the creation of custom protein databases from RNA-Seq data. This allows for the identification of novel peptides that would be missed by standard proteomic searches, significantly enhancing our ability to discover new biological insights from complex genomic and proteomic datasets.

What Problem Does This Paper Address?

Identifying peptides from mass spectrometry (MS/MS) data typically relies on comparing experimental spectra against a reference protein database. However, this approach is inherently limited: if a peptide's sequence isn't present in the reference database, it simply won't be identified, even if the experimental data is of high quality. This is a major bottleneck for discovering novel proteins, protein variants (like single amino acid polymorphisms or INDELs), and peptides arising from alternative splicing or novel transcripts. While RNA-Seq has revolutionized our ability to understand transcriptomes, integrating this information into proteomic analysis for novel peptide discovery has been challenging due to the lack of comprehensive, user-friendly pipelines. Existing tools often focus on only one aspect (e.g., database construction or specific variant types) but not an end-to-end solution.

Methods

PGA provides a comprehensive R/Bioconductor pipeline for novel peptide identification. The core of the method involves two main workflows for constructing customized proteomic databases from RNA-Seq data:

  1. Genome-guided: This workflow uses RNA-Seq data in conjunction with a reference genome. It leverages standard bioinformatics tools (like GATK, SAMtools, TopHat, Cufflinks) to generate variant call format (VCF) files (for SNVs and INDELs), BED files (for splice junctions), and gene transfer format (GTF) files (for novel transcripts). These are then used to build a database that includes canonical proteins, potential novel peptides from variants, and their corresponding reverse sequences.
  2. De novo assembly: For species lacking a reference genome, PGA can use RNA-Seq data that has been assembled de novo (e.g., using Trinity). The resulting transcript sequences are then translated into potential protein sequences.

Once the customized database is built, PGA integrates with various MS/MS search engines (X!Tandem is the default) to identify peptides. Crucially, it includes a post-processing step that employs a specialized False Discovery Rate (FDR) calculation for novel peptides, acknowledging the increased uncertainty when searching against a non-standard database. Finally, it generates an interactive HTML-based report with visualizations for quality checking and data exploration.

Datasets

The authors evaluated PGA using a publicly available dataset from the Jurkat cell line, which included both RNA-Seq and MS/MS data collected in parallel. The RNA-Seq data was obtained from the NCBI Gene Expression Omnibus (GEO) repository (accession number GSE45428), and the MS/MS data was sourced from the PeptideAtlas repository (PASS00215). This dataset allowed for the comparison of PGA's performance against both a reference database and a database derived from de novo transcriptome assembly.

Key Insights and Ideas

  • Bridging the Genomics-Proteomics Gap: PGA effectively bridges the gap between genomic (RNA-Seq) and proteomic (MS/MS) data, enabling the discovery of peptides that would otherwise remain hidden.
  • Comprehensive Novelty Detection: The package is designed to identify a broad spectrum of novel peptides, including those arising from single nucleotide variants (SNVs/SNPs), insertions/deletions (INDELs), alternative splicing events, and entirely novel transcripts.
  • Platform Independence and Ease of Use: By being an R/Bioconductor package, PGA offers a flexible and accessible solution for researchers already working within this ecosystem, aiming for a user-friendly, end-to-end experience.
  • Importance of Customized Databases: The results highlight that customized databases derived from RNA-Seq can lead to a higher number of identified peptides compared to relying solely on standard reference databases.
  • De Novo Assembly as a Viable Alternative: For species without a reference genome, de novo transcriptome assembly provides a robust foundation for building proteomic databases and identifying novel peptides.

Key Results

When applied to the Jurkat cell line dataset, PGA successfully identified 636 novel peptides. These were categorized as:

  • 510 single amino acid polymorphism (SAP) peptides
  • 2 INDEL peptides
  • 49 splice junction peptides
  • 75 novel transcript-derived peptides

The search score distribution for novel peptides was comparable to that of canonical peptides, suggesting high confidence in the identifications. The number of peptides identified using the customized database (73,443) was slightly higher than that identified using the reference database (72,956). Furthermore, the comparison between the genome-guided and de novo assembly workflows showed a significant overlap (91.71%), indicating that both approaches are complementary and can capture unique information.

Limitations and Future Work

The paper acknowledges that while the customized database approach can increase sensitivity, it also leads to significantly larger database sizes, which can be a computational challenge. The authors suggest that further optimization of database construction and search strategies could address this. Future work could involve expanding the types of novel peptide sources that PGA can handle and further refining the post-processing and visualization capabilities to make the interpretation of complex proteogenomic results even more intuitive.

Summary updated

BibTeX
@article{wen2016pga,
  author = {Bo Wen and Shaohang Xu and Ruo Zhou and Bing Zhang and Xiaojing Wang and Xin Liu and Xun Xu and Siqi Liu},
  title = {{PGA: an R/Bioconductor package for identification of novel peptides using a customized database derived from RNA-Seq}},
  journal = {BMC Bioinformatics},
  year = {2016},
  doi = {10.1186/s12859-016-1133-3},
}
Figure