Output formats

Centrifuger produces two tab-separated files: a per-read classification from centrifuger, and a per-taxon profile from centrifuger-quant. Both go to standard output.

Inputs#

The primary input to Centrifuger is the index of the genome or protein database (-x) together with gzipped or uncompressed FASTQ read files — -1/-2 for paired-end data, -u for single-end, -i for interleaved.

Classification output#

centrifuger writes one row per read assignment, with eight columns:

readID    seqID          taxID     score   2ndBestScore   hitLength   queryLength   numMatches
1_1       MT019531.1     2697049   4225    0              80          80            1

Reading the file#

• A read can appear more than once. With -k greater than 1, near-equal assignments each get their own row; numMatches tells you how many rows to expect for that read.
• score versus 2ndBestScore. A high score with a 2ndBestScore of 0 is an unambiguous assignment. Two nearly equal scores mean the read cannot distinguish the two references — the usual situation for conserved regions and closely related strains.
• hitLength versus queryLength. Their ratio is how much of the read actually matched. A long read with a short hit length is weak evidence even when the score looks large.
• Unclassified reads are not represented by a classification row. To collect them, pass --un to centrifuger.

When barcode and UMI extraction are enabled, those values ride along with the read so that assignments can be grouped by cell afterwards — see single-cell and barcoded data.

Quick summaries#

# how many reads were classified
cut -f1 classification.tsv | tail -n +2 | sort -u | wc -l

# most frequently hit taxa
tail -n +2 classification.tsv | cut -f3 | sort | uniq -c | sort -rn | head

# keep only confident, unambiguous assignments
awk -F'\t' 'NR==1 || ($4 > 300 && $5 == 0)' classification.tsv > confident.tsv

Quantification output#

centrifuger-quant estimates the abundance of each taxonomy ID. The report has seven columns:

name                                                          taxID    taxRank  genomeSize  numReads  numUniqueReads  abundance
Legionella_pneumophila_subsp._pneumophila_str._Philadelphia_1  272624   strain   3397753     50        48              0.392641

Reading the report#

• numReads against numUniqueReads. A taxon with many reads but few unique ones is hard to distinguish from its neighbours; one with a high unique count is well supported.
• abundance is length-normalised, so it is comparable between a small virus and a large bacterium in a way that raw read counts are not.
• Rows exist at several ranks. Reads that cannot be resolved to strain are attributed to the lowest rank that does explain them, so a profile mixes strain, species and higher-rank rows.

Alternative formats#

--output-format re-lays the same information for downstream tools:

centrifuger-quant -x cfr_idx -c classification.tsv --output-format 3 > kreport.txt

Since these are all derived from the classification file, emitting several costs nothing beyond a second centrifuger-quant run.

Index files#

centrifuger-build writes <prefix>.1.cfr through <prefix>.4.cfr. These four files are the whole index; the reference FASTA and intermediate build files can be deleted once the build succeeds. Pass the prefix — the path without .1.cfr — to -x.

With --checkpoint, the build also writes <prefix>_checkpoint.[123], which exist only to resume an interrupted build and can be removed afterwards.