A bacterial phyla dataset for protein function prediction.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Protein function prediction is a critical challenge in computational biology due to the vast number of uncharacterized proteins.
• A significant gap exists between protein structures and their known functions, hindering biological understanding.
• The rapid influx of new protein sequences necessitates efficient computational methods for function annotation.

Purpose of the Study:

• To develop a comprehensive dataset for protein function prediction using Gene Ontology (GO) Molecular Function terms.
• To facilitate the training and evaluation of deep learning models for accurate protein function annotation.
• To address the challenge of annotating orphan proteins lacking discernible sequence relatives.

Main Methods:

• Extracted and filtered 171,212 reviewed prokaryotic protein sequences from UniProtKB across 9 bacterial phyla.
• Generated 9890 features per protein, including sequence, sub-sequence, physiochemical, and annotation-based attributes.
• Created a stratified training/testing dataset (3:1 ratio) and a separate dataset of unreviewed pathogenic proteins for function prediction.

Main Results:

• Compiled a large-scale dataset with 171,212 samples, 9890 features, and 1739 GO Molecular Function terms.
• The dataset is well-balanced with sufficient positive and negative samples for each GO term, suitable for deep learning.
• A train/test split of 128,409 and 42,803 samples respectively was established for model validation.

Conclusions:

• The curated dataset provides a robust resource for advancing protein function prediction methodologies, particularly deep learning approaches.
• This resource will aid in uncovering the functions of uncharacterized proteins, potentially revealing key roles in biological processes.
• The dataset supports the development of more accurate and efficient computational tools for bioinformatics research.