Author Correction: Cross-phyla protein annotation by structural prediction and alignment

Fabian Ruperti, Νικόλαος Παπαδόπουλος, Jacob M. Musser, Milot Mirdita, Martin Steinegger, Detlev ArendtGenome 2024 · DOI · OpenAlex

Introduction

In 2023, the research paper "Cross-phyla protein annotation by structural prediction and alignment" was published in Genome Biology (24, 113). This study represented a significant advancement in bioinformatics, proposing a method to annotate the "dark matter" of the proteome—proteins whose sequences have diverged so significantly over evolutionary time that traditional sequence-based tools like BLAST fail to identify them. By leveraging 3D structural conservation through tools like AlphaFold and structural alignment algorithms, the authors demonstrated that protein function and homology could be mapped across distant phyla.

To maintain the integrity and reproducibility of this work, the authors have issued a formal correction addressing two specific bibliographic and data-linking errors. This update ensures that the scientific record remains accurate and that the data supporting the study's conclusions is fully accessible.

Addressing the "Dark Matter" of the Proteome

The original paper tackled a fundamental challenge in evolutionary biology: identifying homologous proteins when genetic sequences no longer show detectable similarity. While DNA and protein sequences change rapidly, the three-dimensional "folded shapes" of proteins often remain conserved over hundreds of millions of years. By shifting the focus from comparing "letters" in a sequence to comparing these folded shapes, the researchers were able to bridge the gap between highly divergent species.

However, the utility of such high-impact bioinformatics research depends heavily on the accuracy of its data provenance. This correction addresses two specific issues that could have hindered the validation and extension of the original work.

Details of the Correction

1. Data Provenance for Spongilla lacustris

A primary component of the original study involved the analysis of protein structures from the freshwater sponge Spongilla lacustris. Sponges are among the most ancient animal lineages, making them critical for understanding the early evolution of multicellularity. The original publication contained an incorrect link to the predicted protein structures for this species, which could have prevented other researchers from building upon the sponge dataset.

The corrected DOI for the S. lacustris structural dataset is now properly identified as: https://doi.org/10.5452/ma-coffe-slac. This update ensures that the specific structural models used to draw the paper's conclusions are findable and accessible.

2. Standardizing the ModelArchive Citation

The original article misnamed and incorrectly cited the "ModelArchive," a vital repository for computational protein structure models that are not derived from experimental X-ray or NMR data. The correction replaces the erroneous "Biological Structure Model Archive" with the standardized citation for the ModelArchive.

Specifically, the text now correctly references the foundational 2009 Structure paper by Schwede et al., which established the standards for protein model applications in research. This correction is significant because it properly attributes the infrastructure that hosts the "structure-first" annotations central to the paper's methodology, reinforcing the importance of institutional support for computational repositories.

Significance for FAIR Data Principles

While these corrections may seem minor, they are essential for upholding FAIR (Findable, Accessible, Interoperable, and Reusable) data principles. In large-scale bioinformatics, a single digit error in a DOI or a misnamed database can "orphan" a dataset, effectively making it invisible to the global scientific community.

As the field moves toward a future where predicted structures become as common as experimental ones, citing the correct infrastructure is essential for data longevity. The correction highlights a broader challenge in bioinformatics: the management of "link rot" and the necessity of adhering to strict database naming conventions. Future work in the field will likely require more automated systems to link structural models to publications to prevent such manual citation errors.

Conclusion

The fundamental findings of the 2023 paper remain unchanged. Structural alignment continues to show superior performance over sequence alignment in identifying homologous proteins across diverse phyla. This correction serves to restore the bibliographic accuracy of the record, ensuring that the "dark matter" of the proteome remains illuminated by accessible and correctly cited data.

Summary updated

BibTeX
@inproceedings{ruperti2024author,
  author = {Fabian Ruperti and Νικόλαος Παπαδόπουλος and Jacob M. Musser and Milot Mirdita and Martin Steinegger and Detlev Arendt},
  title = {{Author Correction: Cross-phyla protein annotation by structural prediction and alignment}},
  booktitle = {Genome},
  year = {2024},
  doi = {10.1186/s13059-024-03196-9},
}
Figure