Species identification

Currently the species identification consists of the information in the ENA metadata, and from running sylph on the raw reads.

Important: the method used to call species from sylph results was made more stringent in release 2025-05. Details are below. If you want a consistent set of species calls for everything up to and including 2025-05, use the 2025-05 aggregated files. For reproducibility, the older (r0.2 and 2024-08) files were not changed and contain the old method species calls.

ENA species

This is simply the value of the scientific_name in the ENA metadata, which is from the original submitter. It uses the NCBI taxonomy.

Sylph species

This uses the GTDB taxonomy, which is inconsistent with the ENA/NCBI taxonomy.

Sylph calls were filtered for quality using the following information:

• c = the effective coverage Eff_cof column from Sylph
• k = kmer length used by Sylph, which for AllTheBacteria was the default 31
• g = reference genome size, which was obtained from GTDB metadata. We used column genome_size from https://data.ace.uq.edu.au/public/gtdb/data/releases/release214/214.0/bac120_metadata_r214.tar.gz and https://data.ace.uq.edu.au/public/gtdb/data/releases/release214/214.1/ar53_metadata_r214.tsv.gz
• r: read length, which was calculated by dividing the ENA base_count by (2 * read_count). Note that base_count is the total of bases across both paired FASTQ files, and read_count is the number of reads in one of the FASTQ files.

To pass, the Sylph call must have:

1. Sequence_abundance at least 99.
2. c * g * (r / (r - k + 1)) >= 0.5 * base_count, which roughly means that at least half of the reads match this genome. Although this may sound like a low threshold, sequencing errors have a significant detrimental affect on the calculation.

Further, for a sample (not just a run) to pass, there can be only one species call that passes. A sample with more than one run could have two different species called, in which case it is failed. If all its passing run calls have the same species, then the sample passes.

2025-05 vs pre-2025-05 calls

Important: only condition 1 above was used to make species calls for releases r0.2 and 2024-08. This means files from 2024-08 and earlier have the calls using only condition 1. The stricter condition 2 was added from 2025-05 onwards. Files from 2025-05 onwards use the stricter condition 1 plus 2. Old files were not changed. This means that to get a consistent set of species calls, use the 2025-05 aggregated files (or the 2025-05 database). There are samples in r0.2/2024-08 that passed condition 1 but not condition 2, meaning that they had a species in r0.2/2024-08 but in 2025-05 are labelled as “unknown”.