Too little reads mapped back to contigs using Bowtie

[oliviaxinw]

Hi,

I assembled Illumina 100bp reads with Velvet after quality trimming (so all the reads are pretty high quality). I used bowtie to map the reads back to the contigs using paired-end option with a maximum mismatch of 3.

Only 35% of the reads mapped to contigs no matter how large a insert size limit I set.
I tried mapping the reads as single reads rather than paired-end and 50% reads mapped. This is still lower than the percentage of reads used to make the contigs.

What could be the problem with this mapping?

[westerman]

Will I be better off mapping the reads not as read-pairs but as individual reads?

Yes. At least then you should be able to tell if all of your reads do map.

[oliviaxinw]

I mapped read 1 as non-paired end reads and 50% read mapped to contigs. This is still less than the amount of reads that are used to make the contigs.

[upendra_35]

I mapped read 1 as non-paired end reads and 50% read mapped to contigs. This is still less than the amount of reads that are used to make the contigs.

My gut feeling is that the k-mer size you chose to build your assembly is not efficient enough. Perhaps you should optimize your k-mer size to efficiently use most of your reads. Hope this helps....

[arolfe]

Do you know how big the target genome was? What's the total size of your contigs? Maybe your assembly didn't cover the whole genome. Or it's possible that you sequenced a lot of reads from some kind of contaminant or artifact that won't map properly.

Have bowtie give you a file of reads that didn't map, then try two things:
1) is there a lot of duplication in that file? eg, maybe most of the unmapped 50% is some artifact like adapter-dimers
2) use blast on a subset of the unmapped reads to figure out whether this is a Bowtie problem (eg, indel in read prevents mapping) or whether the reads really have no overlap with your assembly.

Alex