I am trying to map paired-end reads from an amplicon sequencing using Bismark. The command line is

bismark --un --ambiguous --maxins 1000 --non_directional --score_min L,-0.6,-1 bismark-genome-folder -1 R1.fastq.gz -2 R2.fastq.gz

This is running extremely slow @ around 50000 reads / hour / thread (an estimate from bam-file). I've checked the mapping on three different machines having different specs using: 1) single thread 2) 4 threads and 3) 10 threads. The reaultant speed I'm getting is similar.

I had run the same mapping in single end mode in past (for R1 and R2 separately) and it ran at a reasonable speed. I don't remember the exact speed but at least it was finished in 1-2 days, whereas I'm still waiting for the results of mapping after 3 days (single thread) or 1 day (10 threads).

Next, I did a quick check to see the differences in time by using first 1000 reads of the same (as above) sample

single-end

time bismark --un --ambiguous --non_directional --score_min L,-0.6,-1 bismark-genome-folder  1kr1.fq

real    0m7.883s
user    0m2.523s
sys 0m0.101s

paired-end

time bismark --un --ambiguous --non_directional --score_min L,-0.6,-1  bismark-genome-folder  -1 1kr1.fq -2 1kr2.fq

real    1m25.260s
user    0m8.001s
sys 0m0.141s

This suggests that paired-end mapping is running at least 10 times slower! Has anyone faced similar issues? How to make the mapping faster?

Finer Details: My reference sequences (total ~250) are almost the same (excluding difference of around 30-40 bp in the middle of sequences), length around 800bp each. 150bp, untrimmed reads.

PS: Cross posted to bismark developers https://github.com/FelixKrueger/Bismark/issues/106