umgap prot2kmer2lca
Maps all k-mers from a FASTA stream of peptides to taxon IDs.

The input is given in a FASTA format on standard input, with a single peptide per FASTA header, which may be hardwrapped with newlines. All overlapping k-mers in these peptides (k configurable via the -k option, and 9 by default) are searched for in the index (as build by the umgap buildindex command) passed as argument. The results are printed on standard output in FASTA format.

$ cat input.fa
>header1
DAIGDVAKAYKKAG*S
$ umgap prot2kmer2lca -k9 uniprot-2020-04-9mer.index < input.fa
>header1
571525
571525
6920
6920
1
6920

Add the -o option to print out 0 for k-mers not found in the index.

$ umgap prot2kmer2lca -o uniprot-2020-04-9mer.index < input.fa
>header1
571525
571525
6920
6920
1
6920
0
0

This command also allows an alternative mode of operation. When memory mapped, it can take some time for the index to be searched. With the -m flag, the complete index will be loaded in memory before operation. This, too, takes some time, but for a single large analysis, this impact is irrelevant compared to the time of analysis. When processing many short files, the index would need to be loaded again and again. Instead of using this command as part of a pipeline, ... | umgap prot2kmer2lca index | ..., it can run in a separate (and persistent) process, reusing the same loaded index. Run umgap prot2kmer2lca -m -s umgap-socket index as a service, and when the index is loaded, change your original pipeline(s) to communicate with the socket using OpenBSD's netcat: ... | nc -NU /path/to/umgap-socket | ....

-m / --in-memory

Load index in memory instead of memory mapping the file contents. This makes querying significantly faster, but requires some initialization time.

-h / --help

Prints help information

-o / --one-on-one

Map unknown sequences to 0 instead of ignoring them

-V / --version

Prints version information

-c / --chunksize c

Number of reads grouped into one chunk. Bigger chunks decrease the overhead caused by multithreading. Because the output order is not necessarily the same as the input order, having a chunk size which is a multiple of 12 (all 6 translations multiplied by the two paired-end reads) will keep FASTA records that originate from the same reads together [default: 240]

-l / --length l

The length of the k-mers in the index [default: 9]

-s / --socket s

Instead of reading from stdin and writing to stdout, create a Unix socket to communicate with using OpenBSD's netcat (nc -NU socket). This is especially useful in combination with the --in-memory flag: you only have to load the index in memory once, after which you can query it without having the loading time overhead each time