Kun-peng enables scalable and accurate pan-domain metagenomic classification

Area of Science:

• Metagenomics
• Bioinformatics
• Computational Biology

Background:

• Metagenomic classification faces memory and runtime challenges due to expanding reference databases.
• Existing tools like Kraken2 struggle with the scale and efficiency required for pan-domain analysis.

Purpose of the Study:

• To introduce Kun-peng, a novel taxonomic classifier designed for ultra-scalable and memory-efficient pan-domain profiling.
• To overcome the memory bottleneck in building and querying large metagenomic reference databases.

Main Methods:

• Developed Kun-peng utilizing an intelligent block-partitioned database structure and optimized search strategies.
• Evaluated Kun-peng on the Critical Assessment of Metagenome Interpretation II benchmark and real-world environmental samples.
• Compared performance against Kraken2, Centrifuger, KrakenUniq, and Sylph in terms of memory, speed, and accuracy.

Main Results:

• Kun-peng reduced memory usage by up to 24-fold and accelerated classification by up to 4.73-fold compared to Kraken2.
• Achieved competitive accuracy with fewer false positives and high sensitivity across diverse datasets.
• Demonstrated significant memory reduction (54-473 fold) and speedup (up to 46-fold) in real-world evaluations on large databases and numerous samples.

Conclusions:

• Kun-peng effectively eliminates memory bottlenecks in pan-domain database construction and classification.
• Enables rapid, scalable, and memory-efficient taxonomic analysis of complex metagenomic datasets.
• Offers improved reference coverage and classification efficiency for environmental, ecological, and exposomic studies.