Different TopHat result between entire and subsampled fastq file

[ConcettaDe4]

Hi all!
I am performing a genome-guided transcriptome assembly with Stringtie.

Before the transcriptome assembly, I have mapped my paired-end reads of the same sample TopHat2 and I performed the transcriptome assembly. In my transcriptome assembly I noticed that in a region, Stingtie assembled two transcripts with TopHat mapped reads. When I check this region, TopHat mapped only one mate of the fragment and the other mate is unmapped, so I obtained two different isoforms .

Subsequently I performed the same analysis with a subset of reads of the same sample. I noticed that TopHat aligned in the same region both mate of the same fragment obtaining only one isoform with splitted reads in the region where before there were no reads.

How it can be explained that when I performed the analysis on a subset of reads TopHat is able to map both mates and using the whole fastq file TopHat did not map both mate in the same region.

Thank you,

Concetta