Protein Sequences Clustered at 90 Percent Identity

85 million protein sequences were generated by clustering 400 million sequences from the Open Genomic dataset at 90% identity and 90% coverage using MMseqs2 linclust. The dataset, created by tattabio, provides a non-redundant protein sequence resource for bioinformatics research.

Use Cases

• Train protein language models on 85 million clustered sequences to learn general sequence representations.
• Benchmark remote homology detection methods using clusters defined by 90% sequence identity.
• Analyze sequence diversity and family size distributions across the clustered protein space.
• Develop and evaluate novel protein function prediction algorithms on this large-scale clustered data.

Strengths

• Contains 85 million protein sequences, providing substantial scale for training.
• Derived from clustering 400 million original sequences, ensuring a broad source.
• Clustered at stringent 90% sequence identity and 90% coverage thresholds to reduce redundancy.
• Created using the established MMseqs2 linclust algorithm for reliable clustering.

Limitations

• Lacks explicit column descriptions or metadata details beyond sequence data.
• Unknown if the dataset includes functional annotations or structural information.
• Derived from a specific source (Open Genomic dataset), which may introduce source-specific biases.

Provenance

Source

Clustered from the Open Genomic (OG) dataset.

Collection Method

MMseqs2 linclust applied to 400M protein sequences at 90% sequence identity and coverage.

Time Range

null

Freshness

Last updated November 2024.

Geography

null