I have not had a chance to try PEAR.
I assume some people would like to merge PE reads into SE reads when most of the inset size is less than 2*read length.
1) How PEAR treat un-overlapped PE reads? discard them?
2) What kind of advantage PEAR brings? I imagine the mapping accuracy won't change much. Structural variants are less accurately identified without pairing information. I can't imagine anything good out of this implementation...