NucleoSeeker-precision filtering of RNA databases to curate high-quality datasets.

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Biology

Background:

• RNA structure prediction is a significant challenge due to limited, variable-quality annotated data.
• Existing deep learning models for RNA structure prediction are hampered by data redundancy, leakage, and poor quality.
• Computational methods complement experimental approaches for biomolecular structure determination.

Purpose of the Study:

• To present NucleoSeeker, a unified framework for curating high-quality, tailored RNA structure datasets from the Protein Data Bank (PDB).
• To provide researchers with granular control over data curation through structure, sequence, and annotation filters.
• To demonstrate the utility of NucleoSeeker in creating nonredundant datasets for assessing RNA structure prediction tools.

Main Methods:

• NucleoSeeker integrates multiple tools into a streamlined data curation framework.
• The tool employs filters at the structure, sequence, and annotation levels for precise dataset tailoring.
• A nonredundant RNA structure dataset was generated using NucleoSeeker to evaluate AlphaFold3.

Main Results:

• NucleoSeeker effectively curates high-quality, nonredundant RNA structure datasets.
• The framework offers researchers extensive control over the data selection process.
• Demonstrated the creation of a tailored dataset suitable for benchmarking RNA structure prediction algorithms.

Conclusions:

• NucleoSeeker significantly enhances the quality of RNA structure datasets.
• The tool is user-friendly, flexible, and valuable for training and evaluating RNA structure prediction methods.
• Facilitates the development and assessment of advanced computational tools in RNA bioinformatics.