Selecting best RNA-seq mapping with TopHat

Question

[mjoyraj]

After mapping my PE RNA-seq with Tophat with three different options (given below). I found the following mapping summary given below in table.

1) Without reference annotation:

tophat -p 8 -r 50 -o "output" "indexed_genome_file" R1.fastq R2.fastq

2) With reference annotation:

tophat -p 8 -G “genes.gtf” -o “tophat_RABM” “Genome” R1.fastq R2.fastq

3) With reference annotation disabling novel junctions:

Tophat --no-novel-juncs -p 8 -G “genes.gtf” -o “tophat_RABM” “Genome” R1.fastq R2.fastq

			Mapping reads to genome with TopHat
			With reference annotation	With reference annotation disabling novel junctions	Without reference annotation
Left reads
	Input		71926313	71926313	71926313
	Mapped		62199375 (86.5% of input)	60663645 (84.3% of input)	61835864 (86.0% of input)
		Multiple alignment	10352306 (16.6%)(477254 have >20)	11540865 (19.0%)(508565 have >20)	15034571 (24.3%) (665149 have >20)
Right reads
	Input		71926313	71926313	71926313
	Mapped		62071170 (86.3% of input)	60575371 (84.2% of input)	61694450 (85.8% of input)
		Multiple alignment	10352990 (16.7%)(477253 have >20	11553883 (19.1%)(508573 have >20)	15030545 (24.4%) (665010 have >20)
Overall mapping rate			86.40%	84.30%	85.90%
	Aligned pairs		57244789	55041529	56591033
		Multiple alignment	9527350 (16.6%)	10609333 (19.3%)	13776265 (24.3%)
		Discordant alignment	4048217 ( 7.1%)	4044795 ( 7.3%)	3755391 ( 6.6%)
		Concordant alignment	74.00%	70.90%	73.50%
No. of junctions			144075	97906	140296

 

Accordingly, I thought “with reference annotation” is the best one. But when I viewed the BAM file with junctions, I found there is lot of junctions with high depth between very distantly located genes. My genes of interest are duplicate genes. I guess pre-filtering the mapping along with some other arguments will further improve the mapping, so I thought of running the mapping with the following options:

tophat -p 8 -G genes.gtf -o SRX528281_tophat_RABM_Prefilter --no-mixed --no-discordant --max-multihits 2 --prefilter-multihits --read-realign-edit-dist 0 Genome R1.fastq R2.fastq

Whether my approach is correct…?? Whether the options included will improve the mapping without excluding important information’s..??  Any suggestion will be highly appreciated….

[GouthamAtla]

I think think you are just making it complicated. If you have a GTF file, just use it. If you are not interested in novel transcripts, disable it.

Anyway these quantitative changes exists even if you run the tools with same set of options multiple times.

[mjoyraj]

Some of the duplicate genes have high similarity or highly similar sequence pattern. So I think, if I do not filter the multiple hits, some false positive novel junctions will be revealed with significance. Though I have a GTF file, the genes on which I am interested are mostly predicted. So from the RNA seq results I am trying to re-annotate the genes and also looking for if the genes have isoforms...