bwa outputs bam with less reads than in input fastq

[seru]

I downloaded a publicly available bam file and ran samtools flagstat on it. The output is here:

482805326 + 0 in total (QC-passed reads + QC-failed reads)
0 + 0 duplicates
235328156 + 0 mapped (48.74%:nan%)
482805326 + 0 paired in sequencing
241402663 + 0 read1
241402663 + 0 read2
235328156 + 0 properly paired (48.74%:nan%)
235328156 + 0 with itself and mate mapped
0 + 0 singletons (0.00%:nan%)
0 + 0 with mate mapped to a different chr
0 + 0 with mate mapped to a different chr (mapQ>=5)

Next, I extracted reads from the bam into two fasta files. The sum of reads in both was 482805326, so equal as in the bam. I aligned the fastq files again (bwa aln and bwa sampe), and ran samtools flagstat on the resulting bam. Here is the output:

387280125 + 0 in total (QC-passed reads + QC-failed reads)
0 + 0 duplicates
387280125 + 0 mapped (100.00%:nan%)
387280125 + 0 paired in sequencing
194031131 + 0 read1
193248994 + 0 read2
375578347 + 0 properly paired (96.98%:nan%)
381759377 + 0 with itself and mate mapped
5520748 + 0 singletons (1.43%:nan%)
5108005 + 0 with mate mapped to a different chr
4124551 + 0 with mate mapped to a different chr (mapQ>=5)

And what I don't understand is where is almost 100M reads gone? I haven't found any information on bwa removing unmapped/bad quality/duplicate reads, so I am wondering. Any hints?

I am using bwa 0.5.9-r16

[Guest]

Same here

I have this same problem. Did you ever figure out how to fix it?

Update: The missing reads in the bam file were due to the following error:
[bwa_sai2sam_pe_core] print alignments... [E::sam_parse1] CIGAR and query sequence are of different length
[W::sam_read1] parse error at line 11261615
[main_samview] truncated file.

All of this happened after I switched to a new reference genome (mm10). I'm not getting the error with mm9. I don't have a good explanation but I think that the version of mm10 I tried to use must have been somehow corrupted.

[dpryan]

Note the "100%" mapped metric, which means that the unmapped reads were excluded from the output.