samtools depth vs mpileup output question

[Kennels]

Hello

Regarding the number of reads covering a position in a .bam alignment file, my understanding is that 'samtools depth' and 'samtools mpileup' output should represent the same thing. From what I've read up their respective explanations are :

depth: " compute the per-base depth"
mpileup: "the number of reads covering the site"


However, I am getting different numbers when these options are run on the same .bam file (which is a paired end alignment file).

For example looking at the last 5 lines of each:

Depth:
samtools depth aln.bam | tail -n 5
samp2 5819 46
samp2 5820 46
samp2 5821 45
samp2 5822 43
samp2 5823 43

mpileup (omitting the last fields):
$ samtools mpileup -f ../samp2.fa aln.bam | tail -n 5
[mpileup] 1 samples in 1 input files
<mpileup> Set max per-file depth to 8000
samp2 5819 C 24
samp2 5820 A 24
samp2 5821 T 24
samp2 5822 G 23
samp2 5823 A 23


Probably a simple reason I am missing. Would someone be able to clarify?

Thanks

[Kennels]

can anyone advice? I've also contacted the samtools mail list, to no reply.
It is probably a very simple misinterpretation on my part.

Is the coverage from mpileup based on how many reads start at that
position while depth is cumulative?

[colindaven]

Ouch. What does it give you in VCF format ?

Perhaps the difference is due to the number of high quality Q>13 bases, although it doesn't seem likely given the difference ? Is checking the alignment and exporting a Pileup file with read qualitiies helpful ?

[Kennels]

After some piecing together of info, when I use the -A option of mpileup, then the numbers match up.
Basically it seems 'samtools depth' uses all reads including anomalous pairs to calculate the depth, whereas 'samtools mpileup' filters out some reads including anomalous pairs. Using 'samtools mpileup -A' makes it include the anomalous pairs.

From http://biostar.stackexchange.com/que...-all-the-reads:

'SAMtools pileup discards unmapped reads, secondary alignments and duplicates. It uses non-unique reads.' (assuming this also applies to mpileup)