What's the acceptable solution to improve sensibility for short reads mapping

[xinwu]

Hi All,
I am here to look for acceptable solutions for my goals of short reads mapping.
Goal 1: mapping short reads as fast as possible
Goal 2: if number of mapped reads is low, improve it
Goal 3: try to eliminate false positive mapping

For goal 1, I've evaluated some tools like Bowtie, BWA, BFAST, SSAHA2,etc. Bowtie wins for the speed and robust. BWA and BFAST report errors.
But the mapped result did not satisfy me with around 20% mapped to transcriptome.
To improve the number of mapped reads. The methods I though are:
Method 1: make QC and cut off 3 end basepairs (I cut off 25 of 100 according to QC result). It can improve ~4%, totally ~24%
Method 2: adjust Bowtie parameters, from v=2 to v=3. It can only improve ~1% ;use y option, but cost 2 times of v=2
Method 3: use a more sensitive tool to remap unmapped reads.Say, use bfast or smalt or maq.

One considering is how to handle false positive mapping?
I wonder what's your solutions for this kind of issue?
Thanks.

[rwenang]

It depends on what is your definition of false positive mapping. Due to repeats, some reads can be mapped into several locations with the same level of quality. One way is to employ some sort of scoring function which (mainly) uses quality value to make an informed decision. CMIIW but im under the impression that the only way to resolve this is to use longer reads.

Anyway, you mentioned transcriptome, what data is this?