Kmer content vs overrepresented sequences

[Saeideh]

What is the difference between Kmer content and overrepresented sequences in fasqc?

[westerman]

Well, the manual says ...

"The analysis of overrepresented sequences will spot an increase in any exactly duplicated sequences, but there are a different subset of problems where it will not work.

If you have very long sequences with poor sequence quality then random sequencing errors will dramatically reduce the counts for exactly duplicated sequences.
If you have a partial sequence which is appearing at a variety of places within your sequence then this won't be seen either by the per base content plot or the duplicate sequence analysis.
The Kmer module starts from the assumption that any small fragment of sequence should not have a positional bias in its apearance within a diverse library. There may be biological reasons why certain Kmers are enriched or depleted overall, but these biases should affect all positions within a sequence equally. This module therefore measures the number of each 7-mer at each position in your library and then uses a binomial test to look for significant deviations from an even coverage at all positions. Any Kmers with positionally biased enrichment are reported. The top 6 most biased Kmer are additionally plotted to show their distribution."

[Saeideh]

I seeee. but still there is a confusion for me. Based on my understanding each position in a read is allocated by one nucleotide. so what does it mean by "measures the number of each 7-mer at each position" ?

For example in position number 5 (in Kmer content graph) there is a pick for CGCCG. What does it mean? My knowledge is very basic and I think position 5 should be allocated only by any of A,C,G or T.

[westerman]

I seeee. but still there is a confusion for me. Based on my understanding each position in a read is allocated by one nucleotide. so what does it mean by "measures the number of each 7-mer at each position" ?

For example in position number 5 (in Kmer content graph) there is a pick for CGCCG. What does it mean? My knowledge is very basic and I think position 5 should be allocated only by any of A,C,G or T.

I am not sure what you mean by the word 'allocated'. We may be having difficulties with English. So I can not answer your question directly. However I can try to explain what is happening, at least as far as I understand it.

Take a read. At base (position) #1 it will have 1 of the possible 16,384 7-mers. At position #2 it will have a (most likely) different of the 16,384 7-mers. And so on for the entire read.

Do the same for all of the other reads.

Then look at position #1. Are any of the 16,384 7-mers found statistically more often at position #1 than at the other bases? If so report it. Ditto for all of the other positions -- report any 7-mers which statistically are found more often at that given position than any of the other positions?

[HESmith]

For example in position number 5 (in Kmer content graph) there is a pick for CGCCG. What does it mean? My knowledge is very basic and I think position 5 should be allocated only by any of A,C,G or T.

It means the kmer BEGINNING at position 5 (i.e., 5=C, 6=G, 7=C, 8=C, 9=G).

[Saeideh]

Am I right?

I am not sure what you mean by the word 'allocated'. We may be having difficulties with English. So I can not answer your question directly. However I can try to explain what is happening, at least as far as I understand it.

Take a read. At base (position) #1 it will have 1 of the possible 16,384 7-mers. At position #2 it will have a (most likely) different of the 16,384 7-mers. And so on for the entire read.

Do the same for all of the other reads.

Then look at position #1. Are any of the 16,384 7-mers found statistically more often at position #1 than at the other bases? If so report it. Ditto for all of the other positions -- report any 7-mers which statistically are found more often at that given position than any of the other positions?

------------------------
Rick based on what you said I get this: In each position, there might be 7 bases (continuously) which are repeated more. like this:

Read1: ACGGTCGGTCG
Read2: GTACCTGTAGC
Read3: CGGTGCTGGTC
Read4: CGTTAGCTTCG
Read5: CGTAAGCTTGC
Read6: CGTGGACGGAT
Read7: GGGTCGGCTTA
Read8: TTTTTCGTCGC
Read9: CTGAGTTGGGC
Read10: ACGCCCGGTCG
Read11: GTACCTGTAGC
Read12: CGGTGCTGGTC
Read:13 CTTTAGCTTCG
Read14: CGTAAGAATGC
Read15: CGTGGACGGAT
Read16: GCGTCTATTAA

In position one of these 16 reads "AGCCCCG" is repeated.

Is it right?

[dpryan]

You're misunderstanding Rick. Let's take the example of ACGGTCGGTCG. Its 7-mers are:

ACGGTCG     position 1
 CGGTCGG    position 2
  GGTCGGT   position 3
   GTCGGTC  position 4
    TCGGTCG position 5

Every other read will have similar 7-mers, again, one starting at each position. The whole point of this is to see if you have a bias of some sequence at a given position.

[Saeideh]

You're misunderstanding Rick. Let's take the example of ACGGTCGGTCG. Its 7-mers are:

ACGGTCG     position 1
 CGGTCGG    position 2
  GGTCGGT   position 3
   GTCGGTC  position 4
    TCGGTCG position 5

Every other read will have similar 7-mers, again, one starting at each position. The whole point of this is to see if you have a bias of some sequence at a given position.

--------
Thank you Devon

Now I know what is Kmer.

I attached Kmer content graph of my output from fastqc. Would you please guide me on how to relate it with your explanation about Kmer? (For example, what are the picks in the graphs? Or in the 2nd position, I have G, but why one in upper and the other is lower in the graph?)

[dpryan]

Honestly, that's just recapitulating the "Per sequence GC content" results, which probably indicate an abundance of high-GC content reads. The graphing method is actually a bit weird in my opinion. It graphs the observed/expected ratio of the given kmers at each position. However, it scales everything so that the range is 0-100. The graph isn't flat because there's always going to be variability in things like this.

As I said, in your case the graph doesn't mean much of anything. In other cases, people sometimes observe interesting patterns in graphs like this. For example, one of the bench scientists I work with was recently looking for the binding sites (and hopefully the accompanying motif) of a particular protein. I could actually see the motif in FastQC graphs like this (well, the per-base sequence content was more useful, but the motif popped up in this graph too).

[Saeideh]

I can't understand the graph.

If someone give me a kmer content graph, I can't analyze the graph. I just can say what kmers are repeated many times based on the table beside the graph