Removing near-neighbour redundancy from large protein sequence collections

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Sequence databases are rapidly growing and contain redundant information.
• This redundancy strains computational resources and reduces the quality of annotations.
• Efficient homology searching requires up-to-date and non-redundant sequence collections.

Purpose of the Study:

• To address the challenges of large and redundant sequence databases.
• To develop a method for creating a representative subset of sequences with reduced redundancy.
• To improve the efficiency and accuracy of biological discovery through enhanced homology searching.

Main Methods:

• Clustering of highly similar sequences to form a representative set with >90% mutual sequence identity.
• Utilizing deca- and pentapeptide composition filters to reduce the need for explicit sequence alignment.
• Applying the algorithm to a comprehensive union of major sequence databases (e.g., Swissprot, Trembl, Genbank).

Main Results:

• Generated a non-redundant database (nrdb90) with a 46% size reduction (260,000 to 140,000 sequences).
• Achieved the all-against-all comparison for 90% sequence identity in 2 days of CPU time.
• Demonstrated faster homology searches and unified annotation for clustered sequences.

Conclusions:

• The developed method effectively reduces sequence redundancy while preserving biological information.
• The nrdb90 database and associated tools offer significant computational savings for biological research.
• This approach enhances the efficiency of large-scale biological discovery and data analysis.