ProtAttn-QuadNet: Protein-Protein Interaction Pairs with ProtBERT Embeddings

573,661 reviewed protein sequence embeddings from UniProtKB and two labeled datasets for protein-protein interaction prediction. The balanced dataset contains 249,814 protein pairs, and an oversampled version contains 1,082,662 pairs. Authored by Md Shahidul Islam and last updated on 2026-05-02.

Use Cases

• Train attention-based deep learning models for PPI prediction based on ProtBERT-derived sequence embeddings.
• Benchmark PPI prediction algorithms using the provided balanced and oversampled labeled datasets.
• Analyze protein interaction networks based on the large collection of reviewed protein entries.

Strengths

• Includes 573,661 reviewed protein sequence embeddings from the authoritative UniProtKB database.
• Provides a balanced dataset of 249,814 labeled protein pairs and an oversampled dataset of 1,082,662 pairs for model training.
• Dataset size is 5.0 GB, indicating substantial data volume.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the primary embedding file is unknown, which may limit suitability assessment.

Provenance

Source

UniProtKB database

Collection Method

ProtBERT-derived sequence embeddings for reviewed protein entries.

Freshness

Last updated 2026-05-02 08:05:50; freshness should be verified.