Cluster analysis of barcoded DNA - r

Question

I am using barcodes to tag mitochondrial DNA strands previous to PCR. The barcode sequences are not known, but they are 18 nucleotides long and directly proceed a known sequence (either CATCAT or TACTAC). Each DNA molecule will get a unique barcode identifier. Once the molecules undergo PCR, I need to cluster the sequences based on their 18 nucleotide barcode, and then subsequently align the sequences, per barcode.
To use an overly simple example, lets say I have 2 molecules that are going into a PCR reaction:
CATCATBARCODE1SEQUENCE1
TACTACBARCODE2SEQUENCE2
After amplification I have:
CATCATBARCODE1SEQUENCE1
CATCATBARCODE1SEQUENCE1
TACTACBARCODE2SEQUENCE2
TACTACBARCODE2SEQUENCE2
I then want to search the section of sequence at position 6-13 and cluster them based on that window of sequence without changing the rest of the sequence, which would actually just look like what I have above. Then I could perform the alignment on the adjacent sequences.
Any ideas on how I could accomplish this clustering of a window of sequence, without taking into account the rest of the sequence? Thanks.

Answer 1

Overly simplified R code, but seems to do what you ask:
seqs <- c('CATCATBARCODE1SEQUENCE1',
'CATCATBARCODE1SEQUENCE1',
'TACTACBARCODE2SEQUENCE2',
'TACTACBARCODE2SEQUENCE2' )
clusters <- list()
for (seq in seqs) {
barcode <- substr(seq, 7, 14)
if (!is.null(clusters[[barcode]])) {
clusters[[barcode]] <- append(clusters[[barcode]], seq)
} else {
clusters[[barcode]] <- c(seq)
}
}
print(clusters)
prints:
$BARCODE1
[1] "CATCATBARCODE1SEQUENCE1" "CATCATBARCODE1SEQUENCE1"
$BARCODE2
[1] "TACTACBARCODE2SEQUENCE2" "TACTACBARCODE2SEQUENCE2"

Answer 2

Assuming you can already obtain sequences starting like [CATCATBARCODEX] what I would do is just to process it in python. If your sequence starts are not the same then you may need to search for CATCAT and discard those that look to be in the wrong position. There may be some issue if the number of barcodes is very large but I think for something on the order of 100,000 simple methods should work.
Anyways, once you find the CATCAT what I would do is just to build up a dictionary of barcodes and start filtering. Then you can just rip off this first part of the sequences and align using whatever methods (I had a barcode project and using custom genome with bowtie was convenient).
let's say you need to find this sequence instead of just starting with it, in python a solution would be like
my_dict= {}
for seq in seqs:
idx = seq.find("CATCAT")
idx2 = seq.find("TACTAC")
if idx==-1 and idx2==-1:continue
# here you need to consider the location of idx and idx2, both may be present, sequence needs to be long enough etc
barcode = seq[idx+6, idx+6+18]
# you may want to shorten the barcode or encode it to a string
if barcode in my_dict:
my_dict[barcode]=1
else :
my_dict[barcode]+=1;
seq=seq[idx+24:]
In addition to the counting you can 1) append sequences to a fasta file per barcode or 2) assign the barcode as annotation to a large fasta file.
Regardless you probably want to strip down the sequence to simplify the downstream analysis.