Deciphering Cigar

Question

[Leszek]

Hi, I need to get 5'-end position of each read in SAM file. For reads aligned in forward direction it's provided in SAM file (POS). But for reads aligned onto reverse strand the one has to figure it out from the cigar string. You cannot simple add length of the read to left-most position, as reads may be soft-clipped or some indel may occure.

I have written simple python function in order to find right most position of aligned read. Could you have a look and comment on that as I'm not sure if it's correct?

import re

#should nicely separate CIGAR entries
cigar_pat = re.compile(r"\d+[MIDNSHP=X]{1}")

def cigar2end( left,cigar ):
  """Return right-most position of aligned read."""
  #store info about each CIGAR category
  counts={ "M":0, #M 0 alignment match (can be a sequence match or mismatch)
           "I":0, #I 1 insertion to the reference
           "D":0, #D 2 deletion from the reference
           "N":0, #N 3 skipped region from the reference
           "S":0, #S 4 soft clipping (clipped sequences present in SEQ)
           "H":0, #H 5 hard clipping (clipped sequences NOT present in SEQ)
           "P":0, #P 6 padding (silent deletion from padded reference)
           "=":0, #= 7 sequence match
           "X":0, #X 8 sequence mismatch
        }
  #split cigar entries
  for centry in cigar_pat.findall(cigar):
    ccount  = int(centry[:-1])
    csymbol = centry[-1]
    counts[csymbol] = ccount
  #get number of aligned 'reference' bases
  aligned = counts["M"] + counts["D"] + counts["N"] #+ counts["="] + counts["X"]
  right   = left + aligned
  return right

[Sean Davis]

You might consider using pysam for this type of thing.

[Leszek]

thanks, got it:)

[Geparada]

I was writing my own codes to process the CIGAR line in order to extract different information that I needed. But now I has been using Pysam and I save a lot of time because it have very useful functions and transform the CIGAR line into a tuple very easy to process.

Answer 1

[Leszek]

As Sean suggested I use pysam:

import pysam
samfile = pysam.Samfile( "somefile.bam","rb" )
#process all reads in region
for sam in samfile.fetch( ref,start,end ):
    print sam.pos, sam.aend

[Pierre Lindenbaum]

does your code return the same start/end ?

[Leszek]

what do you mean by the same start/end? it returns first and last aligned base of the read on the ref

[Bionic]

Thanks for sharing your code. I used them to find start and end position of reads but it asks me to define ref/start/end. Any suggestions?

[Sean Davis]

You need to provide the values for "ref", "start", and "end". The code, as written above, assumes that has been done elsewhere.

Answer 2

[marcomonti.mm94]

Hi,

I found myself in a situation where I cannot use pysam. Therefore, I have created a python function to calculate the number of bases of each type in the CIGAR string. I started from your function. Maybe it could be useful also for you.

def cigar_to_dict(CIGAR: str):
# store info about each CIGAR category
cigar_dict = {"M": 0,  # M 0 alignment match (can be a sequence match or mismatch)
          "I": 0,  # I 1 insertion to the reference
          "D": 0,  # D 2 deletion from the reference
          "N": 0,  # N 3 skipped region from the reference
          "S": 0,  # S 4 soft clipping (clipped sequences present in SEQ)
          "H": 0,  # H 5 hard clipping (clipped sequences NOT present in SEQ)
          "P": 0,  # P 6 padding (silent deletion from padded reference)
          "=": 0,  # = 7 sequence match
          "X": 0}  # X 8 sequence mismatch

# Note the count variable is a string
count = ""
for i in CIGAR:
    if i.isnumeric():
        count = count + i
    else:
        cigar_dict[i] = cigar_dict[i] + int(count)
        count = ""      # reset count
return cigar_dict

[cmdcolin]

not sure this is correct, I don't know python well but if you are processing the string one character at a time then you will not be handling numbers >=10...

[marcomonti.mm94]

To overcome the problem you mentioned I summed the numbers as strings: count = count + i . For example, the number 17 will be "1" + "7" = "17" and then I converted it in integer int("17"). P.S. I only sum the string if is a number (if i.isnumeric():). It is a bit ugly but it works.

[cmdcolin]

Nice, good fix. that method is actually a bit faster than a regex based approach in a little microbenchmark i whipped up (js, not python) https://jsbench.me/gpl0pyj08n/1