Table 1:

Running times for classifying 9.4 million reads (from a human eye metagenome, accession SRR12486990, available from NCBI at https://www.ncbi.nlm.nih.gov/sra/SRR12486990) with 8 threads using KrakenUniq in different modes. The database size was 392 GB, and it consisted of all complete bacterial, archeal, and viral genomes in RefSeq from 2020, 46 selected eukaryotic human pathogens (Lu & Salzberg, 2018)), as well the human genome and a collection of common vector sequences. The database is available for download at https://benlangmead.github.io/aws-indexes/k2 under the name EuPathDB46. The command lines used to measure the runtimes were krakenuniq --db krakendb-2020–08-16-all_pluseupath --threads 24 --report-file report --output classify SRR12486990.fastq with no additional options for default, and with addition of the preload option shown in the table for various preload sizes. in the database chunking experiments (using –preload-size) KrakenUniq loaded the database into RAM in 49, 25, 13 and 7 chunks (respectively).

Mode	Running time
default (memory mapping)	48 hours
--preload	14 min
--preload-size 8G	47 min
--preload-size 16G	32 min
--preload-size 32G	25 min
--preload-size 64G	19 min