Negative regions from genNullSeqs overlap test regions

[kwcurrin]

Hello,

Thank you for developing this tool.

It appears that genNullSeqs can generate negative sequences whose regions overlap the test set. I generated a test set bed file by taking the top 25,000 macs2 peaks ranked by decreasing fold enrichment. I only took the first 200bp of the peaks for simplicity for this test:
sort -k 7,7nr ${in_file}
| head -25000 | awk '{print $1"\t"$2"\t"$2+200}' > test.bed

I then ran the below R code
library(gkmSVM)
library(BSgenome.Hsapiens.UCSC.hg38.masked)

genNullSeqs("test.bed",genome=BSgenome.Hsapiens.UCSC.hg38.masked,
batchsize=100000,length_match_tol=0)

Strangely, there are hundreds of negative regions that overlap test regions:
intersectBed -u -a negSet.bed -b test.bed
| wc -l

result: 428

I checked with a larger fraction overlap to make sure these overlaps weren't due to edge cases from 0 and 1-based coordinate issues:
intersectBed -u -f 0.5 -a negSet.bed -b test.bed
| wc -l

result: 223

I've also noticed this with a smaller batchsize of 50,000 as well.

Thanks for any help you can provide,

Kevin