I tried a second ago, and got:

> anno.database
GRanges object with 58153 ranges and 1 metadata column:
             seqnames               ranges strand |   gene_name
                <Rle>            <IRanges>  <Rle> | <character>
     DDX11L1     chr1       [11869, 14412]      + |     DDX11L1
      WASH7P     chr1       [14363, 29806]      - |      WASH7P
  MIR1302-10     chr1       [29554, 31109]      + |  MIR1302-10
     FAM138A     chr1       [34554, 36081]      - |     FAM138A
      OR4G4P     chr1       [52473, 54936]      + |      OR4G4P
         ...      ...                  ...    ... .         ...
     CYCSP49     chrY [28695572, 28695890]      + |     CYCSP49
  SLC25A15P1     chrY [28732789, 28737748]      - |  SLC25A15P1
     PARP4P1     chrY [28740998, 28780799]      - |     PARP4P1
     FAM58CP     chrY [28772667, 28773306]      - |     FAM58CP
     CTBP2P1     chrY [59001391, 59001635]      + |     CTBP2P1
  -------
  seqinfo: 24 sequences from GRCh37 genome
> seqlevelsStyle(anno.database) <- "NCBI"
Error in `levels<-.factor`(`*tmp*`, value = c("1", "2", "3", "4", "5",  :
  number of levels differs

  The same error I got when I tried to give values to seqlevels(anno.database) or seqnames(seqinfo(anno.database))

I found what the problem is.

The original GRangres object is:

anno.database = toGRanges(EnsDb.Hsapiens.v75, feature="gene")

if I run 

seqlevelsStyle(anno.database) <- "NCBI"

This works.

But I deleted some rows in the object to keep only Chr1:22, X, Y. If I run seqlevelsStyle after deleting process, it gives that error message in my previous message.

Of course, I can do filtering after changing seqlevels, but curious why the second method should not work? I remember I had problems that stemmed from the same point here. I feel that when I filter rows of GRanges object, I need to make changes on some other information.

Here is how I filter rows for a GRanges object:

idx = which(seqnames(anno.database) %in% c(as.character(1:22),"X","Y"))
anno.database = anno.database[idx, ]
anno.database@seqinfo = seqinfo(anno.database)[c(as.character(1:22),"X","Y"), ]