Subset (reduce) bam to contain only variant containing reads

Question

[SemiQuant]

I have targeted sequencing alignments bam files. I want to "pull out" only read pairs in which the reads have a variant, i.e. do not perfectly match the reference. It doesn't matter if these are sequencing errors etc.

Any suggestions, simply using samtools would be great but I could find any bam flags for this, a pysam solution would also be great.

[GenoMax]

Some past threads to get you going:

How to extract reads with a known variant form a bam file
Extract PE Reads (with their mates) supporting variants in vcf file
Extracting Reads Containing A Specific Variant From A Bam File
Sam/Bam: Extract Only Alignment Reads That Contain A Specific Position
Extract reads with variant at given position

Answer 1

[SemiQuant]

I don't know why Pierre Lindenbaum answer doesn't show but he proposed this solution.

samtools view -h in.bam | grep -E '^@|NM\:i\:[^0]' > out.sam

NM:i:count = Number of differences (mismatches plus inserted and deleted bases) between the sequence and reference...

If it's missing you can add it using

samtools calmd  "$bam"  "$bam"

An alternate solution is to use bamtools like so.

bamtools filter -tag "NM:>0" -in in.bam -out out.bam

you can use >, >=, <, <=, ! in the expression