Miseq assemblies

[RGLADSTONE]

Bacteria sequencing. Nextera XT library prep. Triplicate technical replicates from a single MiSeq run 2 x 250 PE, de novo assembly on MiSeq gives ~450 contigs for all three. Optimized velvet assembly from fastq's gives 200 contigs for two samples and 800 contigs for one. Why would there be so much of a difference between 3 technical replicates for one method of assembly and not for the other?
To make things more complicated the triplicate samples are from a single strain but three individual DNA extractions.

[mwatson]

Bacteria sequencing. Nextera XT library prep. Triplicate technical replicates from a single MiSeq run 2 x 250 PE, de novo assembly on MiSeq gives ~450 contigs for all three. Optimized velvet assembly from fastq's gives 200 contigs for two samples and 800 contigs for one. Why would there be so much of a difference between 3 technical replicates for one method of assembly and not for the other?
To make things more complicated the triplicate samples are from a single strain but three individual DNA extractions.

1) is there a reference? If so, map and then estimate the insert size
2) Try FLASH to see if the reads overlap
3) Trim to 100bp PE and assemble again -> does this help?
4) Randomly choose 10%, 25%, 50% of the data, assemble again, does this help?
5) have you checked for adapters?