Why Is My Uniquely Mapped Read Count in Samtools Higher Than in Bowtie2?

[M.Askari]

Hi everyone,

I aligned my paired-end sequencing data using Bowtie2 and processed the resulting BAM files with Picard MarkDuplicates. However, I noticed a discrepancy between Bowtie2’s uniquely mapped read count and Samtools’ uniquely mapped reads after filtering. Bowtie2 alignment statistics (bowtie2.txt):


8536975 reads; of these: 5180282 (60.68%) aligned concordantly 0 times (unmapped) 1868933 (21.89%) aligned concordantly exactly 1 time (uniquely mapped) 1487760 (17.43%) aligned concordantly >1 times (multi-mapped) Overall alignment rate: 39.32%

• Uniquely mapped reads reported by Bowtie2: 1,868,933

Samtools read count after removing duplicates (samtools view -c -q 30 CR05NFYA_S1_mdu.bam):

2,877,386

• Samtools reports ~1M more uniquely mapped reads than Bowtie2.

Questions:

1. Why does my final _mdu.bam file (uniquely mapped, deduplicated) contain ~1M more reads than Bowtie2’s unique count?
2. Does Bowtie2 apply stricter filtering than samtools view -q 30 when classifying uniquely mapped reads?
3. Could multi-mapped reads or soft-clipped alignments still be counted in _mdu.bam, even with MAPQ ≥ 30?
4. How does paired-end read counting differ between Bowtie2 and Samtools? Does Bowtie2 exclude some properly paired reads?

What I Have Tried:

• Verified read counts before and after duplicate marking:bash
  CopyEdit
  samtools view -c -F 1024 CR05NFYA_S1_mdu.bam
• Checked MAPQ distribution:bash
  CopyEdit
  samtools view CR05NFYA_S1_mdu.bam | awk '{print $5}' | sort -n | uniq -c
• Excluded secondary alignments:bash
  CopyEdit
  samtools view -c -F 256 CR05NFYA_S1_mdu.bam

I would appreciate any insights!

Thanks in advance!

[fchatonnet]

Dear M. Askari, bowtie2 indicates the number of pairs of reads which are correctly aligned exactly once, so the number of uniquely mapped reads after bowtie2 usage is 3,737,866 (1,868,933 * 2).
After removing the duplicates, samtools view -c counts the single reads in the bam file, not the pairs and you end up with 2,877,386 reads, meaning that duplicate filtering eliminated 860,480 duplicate reads.
Cheers and good luck!