Using the same paired-end data, MACS2 BAMPE mode gave mean fragment size of 111bp, while BEDPE mode gave mean fragment size of 259bp.
To produce bedpe files from BAM files, I followed one previous thread in the MACS2 GitHub repository: https://github.com/taoliu/MACS/issues/183
Basically, I used "samtools view -b -f 2 -F 4 -F 8 -F 256 -F 512 -F 2048" to select reads first, and finally cut 3 columns of bedtools ouput to make the bedpe files specific to MACS2.
I also used "Picard collectinsertsizemetrics" to check insert size distribution, the median of which is very close to the "mean fragment size" in BEDPE mode. The size in BAMPE mode seemed mistakenly small.
I am using MACS2 126.96.36.19960309 installed with miniconda.
I also posted the same question (along with two log files) in the MACS2 repository at GitHub ( https://github.com/taoliu/MACS/issues/217 ), but have not gotten any feedback. I guess there are more MACS2 users in Biostars. So I re-post the question here, just to make sure I am using MACS2 correctly.
I noticed mcjmigdal in this thread (MACS2 produces different results using BAMPE and BEDPE) also ran into this problem. Hope we can sort this out.
Many thanks, Xinwei