The original paper actually used a collection of TFs + Mediator. ROSE was not designed for H3K27ac use, though both the authors and others frequently use it. You'd likely get these same comments in review if you tried it. I actually kind of expect what you get out using ATAC would be more believable, as the default promoter exclusion settings from ROSE are rather opaque and too strict in my experience for H3K27ac.
What I mean by that is you get lots of SEs called that are caused by peaks in the promoter region that only get removed from stitching if they are fully encapsulated by the promoter region, which I think ROSE defines as +/- 2kb of the TSS by default. You often get H3K27ac peaks that hit those borders, are included in stitching, and thus included in the enhancer regions ranking.
Regardless, if you're using ROSE to do the SE calling, be sure to use an updated annotation. The ones included are from like 2011 and woefully out of date. And consider expanding your definition of a "promoter" to be a little wider to mitigate instances as described above if using H3K27ac.