How can I detect a split read in a BAM file? Is there any sign in the CIGAR string that describes split read?
The answer to this question depends on the aligner used to generate SAM/BAM file. For BWA method that is part of the pipeline mentioned in comments, split reads will be flagged as supplementary alignment (2048). You can modify this behavior by using option -M as described in the BWA-MEM manual page.
So, in case of split alignment, you will end up with multiple aligned segments that have identical query name and they are flagged as 2048. These segments may only have part of the original read sequence, but the complete read sequence is present in at least one of the segments.
Also, there is no unique sign in CIGAR of an individual segment to identify split reads. Each aligned segment of the original read may have long deletion in its CIGAR or aligned pairs. However, without looking at all segments with the read ID it is not clear whether the segment has a novel sequence in it that doesn't align with reference or that part matches different chromosome/coordinates.