"beginner in alignment" question

[bioinfosm]

I noticed similar behavior looking at capture illumina reads. The capture is not very specific and there is always a lot of 'off-target' sequencing that occurs. In case of using a restricted reference sequence, a lot of those 'off-target' reads are forced to align to these regions, producing false variants.

The best approach is to use the whole genome for the mapping of reads.

[menenuh]

Hello,
I am analyzing Roche 454 sequence data. The sequencing was performed not for whole genome but for the exons of (around) 100 genes. When I first started analyzing, I used the target sequence (exome of 100 genes) as my reference not the whole genome. After all, this target region is tiled and sequenced by Roche platform. But by doing so I am getting a lot of consequential SNPs, most of them are probably false positives.
But when I perform whole genome alignment, I am getting reasonable/low number of SNPs.
I checked some specific regions which are showing great variation at the number of SNPs. Turns out, some of the reads mapping to that region in the 1st alignment are not mapping there in whole genome alignment, but to some other region which is not in the target sequence.
So my questions are;
Is it general practice to do whole genome alignment in any given NGS project, or do I need more stringent alignment parameters when performing alignment over a specific region?
By the way, I am using bwa and ssaha2 for alignment step.

Thanks!