strand is ignored and stranded (seqnames, start, end, strand) tuples are treated as invalid

[ttriche]

R> subset(gsrd, strand == "-")

DataFrame with 12241 rows and 7 columns
             seqnames     start       end      strand         name       idx
                <Rle> <integer> <integer> <character>  <character> <integer>
LOC729737        chr1    140339    139790           -    LOC729737         2
LOC101928626     chr1    629010    585990           - LOC101928626         3
LINC00115        chr1    827522    826206           -    LINC00115         6
FAM41C           chr1    876903    868071           -       FAM41C         7
LINC02593        chr1    919692    916870           -    LINC02593         8
...               ...       ...       ...         ...          ...       ...
CMC4             chrX 155071362 155061622           -         CMC4       948
MTCP1            chrX 155147937 155064034           -        MTCP1       949
RAB39B           chrX 155264589 155258241           -       RAB39B       952
CLIC2            chrX 155334657 155276211           -        CLIC2       953
TMLHE            chrX 155669944 155490115           -        TMLHE       956
                   assay
             <character>
LOC729737     GeneScores
LOC101928626  GeneScores
LINC00115     GeneScores
FAM41C        GeneScores
LINC02593     GeneScores
...                  ...
CMC4          GeneScores
MTCP1         GeneScores
RAB39B        GeneScores
CLIC2         GeneScores
TMLHE         GeneScores
R> makeGRangesFromDataFrame(subset(gsrd, strand == "-"), ignore.strand=TRUE)

Error in .width_as_unnamed_integer(width, msg = "an end that is greater or equal to its start minus one") : 
  each range must have an end that is greater or equal to its start minus one

Enter a frame number, or 0 to exit   

1: makeGRangesFromDataFrame(subset(gsrd, strand == "-"), ignore.strand = TRUE)
2: IRanges(ans_start, ans_end, names = ans_names)
3: new_IRanges(start = start, end = end, width = width)
4: .new_IRanges_from_start_end(start, end)
5: .width_as_unnamed_integer(width, msg = "an end that is greater or equal to 

Selection: 0

R> packageVersion("GenomicRanges") 

[1] '1.62.0'
R> packageVersion("IRanges")

[1] '2.44.0'

[ttriche]

arguably this is notABug but it sure seems like something that could be handled on the fly?

# GRanges issue, sidestep 
gsrd$startpos <- pmin(gsrd$end, gsrd$start)
gsrd$endpos <- pmax(gsrd$start, gsrd$end)
gsrd$start <- gsrd$startpos
gsrd$end <- gsrd$endpos 
gsrd$endpos <- NULL 
gsrd$startpos <- NULL 
gsrd$assay <- NULL

is(as(gsrd, "GRanges"), "GRanges") 
# [1] TRUE 

[hpages]

Fixing the starts and ends with pmin() and pmax() is equivalent to swapping them when end < start. More precisely, it's the same as doing:

fixme <- end < start
fixed_start <- ifelse(fixme, end, start)
fixed_end <- ifelse(fixme, start, end)

Maybe that's the right thing to do in your case but that won't always be the case. For example, if the input data frame is expected to contain 0-width ranges (e.g. insertion points), then some ranges are actually expected to have end == start - 1, and those should not be swapped.

Generally speaking the business of automatically repairing broken input on the fly is tricky and risky. I'd rather fail loudly with an informative error message so the user knows that something went wrong upstream and has a chance to fix it.