So a GRangeslist have group and group_name, but when you do this conversion, the info is lost:

Grangeslist -> data.frame -> GRanges ->GrangesList

Is there a way to preserve the information ?

I tried to save the grouping of the original GRangeslist, but since the final GRangeslist only have 1 group, it does not work.

Of course you can do something like:

#df.gr: a global data.frame of all granges

GRangesList(lapply(df.gr$names, function(x) getGRLbyName(x)))

getGRLbyName = function(name){
  tempEquals = df.gr[df.gr$name == name] #find all in group by name
  a = GRanges( tempEquals )
  names(a) = rep(x,nrow(tempEquals)) # make names
  return(a)
}

But this is just bad..

Another artifact is that now the group will be added as a metacolumn in the final GRangesList.

How do you guys do this ?

Next thing is how GRanges accesses rows:

Usually this is done by name, i.g gr["ENST0009123123"], but what if I want to access by exon_id, which is a meta column ?

It looks like this is not possible ? and I therefor have to transform to data.frame, and I lose the grouping..

I feel like GRanges is not suited if you want to do big changes to specific data in it, because you can not do what you want, unless you transform to data.frame

Information will be lost through coercion to a more general data structure. I guess there could be a way to coerce from data.frame to GRangesList, making an inference on the grouping column based on its name. 

Typically though this round trip should not be necessary. GRanges shares much of the same functionality as data.frame. For example, to subset to a specific exon_id, you could do something like:

gr[gr$exon_id == "exon1"]