How To Handle Ns In The Middle Of Reads

Question

[kautilya]

For my illumina data fastqc shows presence of N's at positions 13,14,15 in 101 bp longs reads. If i go for cropping first 15 bases by using trimmomatic, it solves the problem but i lose a lot of data. I wanted to know that if i retain the N's what sort of problems would they cause during alignment(bwa+stampy)/variant calling(unified genotyper) and how can i handle these problems?

If any body faced a similar problem how did you handle it?
Similar questions asked on different forums but none has answered.
Could not find a resourse on how variant calling programs handle N's. Do they ignore them? Or consider them as a variation with low confidence scores?

Following is the image for per base n content from fastqc
http://i43.tinypic.com/sfyz5z.jpg

[Manu Prestat]

Shouldn't you first investigate why you got those weird Ns at these positions?

[kautilya]

These are possibly due to machine read errors during sequencing. These are particular to only 1 of 3 runs. Looking for a way of handling these without losing a lot of sequence data.

Answer 1

[Gabriel R.]

If you want to do a BWA followed by GATK, I would use your reads as is. They likely have a base quality of 0 and GATK overlooks them. BWA will substitute them for random bases but fallacious alignments induced by those bases will be rare. 

The cause ? If this is Illumina, sometimes the reagents do not make it to the flowcell for a cycle or two due to pump problems or air bubbles.