Why does Cufflinks assign a strand to some RNA sequences but not others?

[gwilymh]

I am analyzing RNASeq data using a Tuxedo pipeline. For the most part, the RNA sequences seem to align with known genes in the correct orientations.

A number of putatively new genes have been discovered, however. In the output GTF file, though, some of these new genes are not assigned to either the sense or the antisense strand, i.e. column 7 contains a "." instead of + or - sign.

Does anyone know why this might be the case? This seems to happen for the new genes but not the already known genes.

In the Cufflinks part of the analysis, I used the --GTF-guide option with a GFF3 file of exon locations that I downloaded from Flybase. I also used
--upper-quartile-norm --min-intron-length 5

[kmcarr]

I am analyzing RNASeq data using a Tuxedo pipeline. For the most part, the RNA sequences seem to align with known genes in the correct orientations.

A number of putatively new genes have been discovered, however. In the output GTF file, though, some of these new genes are not assigned to either the sense or the antisense strand, i.e. column 7 contains a "." instead of + or - sign.

Does anyone know why this might be the case? This seems to happen for the new genes but not the already known genes.

In the Cufflinks part of the analysis, I used the --GTF-guide option with a GFF3 file of exon locations that I downloaded from Flybase. I also used
--upper-quartile-norm --min-intron-length 5

A couple of questions:

Was the RNA-Seq library prepared using a strand-specific prep method?

If it was prepared using a strand-specific method where the proper parameters passed to TopHat and Cufflinks?

If an RNA-Seq library is prepared using a method which does not preserve strand information (e.g. TruSeq RNA Prep Kit v2) then Cufflinks is unable to determine simply from the alignment which strand the original RNA was transcribed from. If the read overlaps a known, annotated gene (i.g. you are providing a GFF file to TopHat/Cufflinks) it will assume the read originated from the known gene for which the strand is already known. Cufflinks creates a novel gene model from a group of aligned reads it can not know a priori which strand the RNA is transcribed from. In some cases, if there enough reads mapping across intron/exon junctions Cufflinks will impute the strand information based on the canonical directionality of the splice junction(s).