using khist to generate kmer coverage histogram

[Brian Bushnell]

The program (conceptually) allocates one counter per kmer, and increments it every time that kmer is seen. So, let's say K=2 and the input string is "CATTATTT".

That breaks down into these kmers:
CA, AT, TT, TA, AT, TT, TT

After reverse-complementing to store only a single canonical copy, either forward or reverse, we get this:

CA, AT, AA, AT, AT, AA, AA

So the kmer counts stored by the program would be:

AA: 3
AT: 3
CA: 1

This would equate to 3 unique kmers. The histogram would look like this:

#Depth	Raw_Count	Unique_Kmers
1	1	1
3	6	2

Generally, (column 3) = (column 2)/(column 1).

So line 1 means there was a single kmer (CA) that occurred exactly once, and it was counted exactly once. Line 2 means that there were 2 unique kmers (AT and AA) that each occurred 3 times, for a total of 6 occurrences.

Therefore - if you want to plot the coverage with respect to the genome, I suggest plotting the "unique" column. And to clarify, the number of "unique kmers" is not the same as the number of kmers that only occur once (I would call those "singleton kmers") - the second number of row 1 gives you the number of singleton kmers counted (1, in this case).

[plumb]

I used khist from bbmap to generate kmer coverage, the attached included the output and the log file. There are a few things I am not very clear. what's different between "Raw count and unique_kmer" in the output file?
Is "unique kmer" defined as "the kmer only appear once in the dataset"? what's the differences between unique kmer and the kmers whose depth is one?
When I plot the dataset for coverage, which column should I use? raw count or unique kmer?

The data is a metagenome data.

Thanks