The documentation of alignmentSieve seems to be wrong.

[ljw20180420]

If, instead, the protocol is strand-specific, then the first set of integers in a pair would be applied to fragments where read 1 precedes read 2, and the second set to cases where read 2 precedes read 1. In this case, the first value in each pair is applied to the end of read 1 and the second to the end of read 2. Take the following command as an example:

.. code:: bash

$ alignmentSieve -b paired_chr2L.bam \
--minFragmentLength 140 --BED \
--shift -5 3 -1 4 -o fragments.bedpe

Given that, the -5 3 set would produce the following::

     ----> read 1
                 read 2 <----

     ------------------------ fragment

-------------------------------- shifted fragment

and the -1 4 set would produce the following::

     ----> read 2
                 read 1 <----

     ------------------------ fragment

         --------------------- shifted fragment

As can be seen, such fragments are considered to be on the - strand, so negative values then shift to the left on its frame of reference (thus, to the right relative to the + strand).

The diagram for -1 4 seems to be wrong. Because -1 and 4 has opposite sign, the shift of read 1 and read 2 should also in the opposite directions (see the shift for -5 3).