Should I trim my MiSeq data?

[cwzkevin]

Hi there,

I have three 151 pe genome MiSeq data for de novo assembly (velvet). Below is the fastqc quality plot of them. (a_1, a_2, b_1, b_2, c_1, c_2)
1. Should I trim them? Or, do you think they are fine?
2. If I should trim them, do I trim /1 to 150 base (remove the last bit) or to 149 base (remove the last two bits), and how about /2?
Thanks in advance!

a_1 and a_2

[cwzkevin]

b_1 and b_2
[ATTACH]1660[/ATTACH]

[ATTACH]1661[/ATTACH]

hi there,

i have three 151 pe genome miseq data for de novo assembly (velvet). Below is the fastqc quality plot of them. (a_1, a_2, b_1, b_2, c_1, c_2)
1. Should i trim them? Or, do you think they are fine?
2. If i should trim them, do i trim /1 to 150 base (remove the last bit) or to 149 base (remove the last two bits), and how about /2?
Thanks in advance!

[cwzkevin]

c_1 and c_2
[ATTACH]1664[/ATTACH]

[ATTACH]1665[/ATTACH]

hi there,

i have three 151 pe genome miseq data for de novo assembly (velvet). Below is the fastqc quality plot of them. (a_1, a_2, b_1, b_2, c_1, c_2)
1. Should i trim them? Or, do you think they are fine?
2. If i should trim them, do i trim /1 to 150 base (remove the last bit) or to 149 base (remove the last two bits), and how about /2?
Thanks in advance!

[Wallysb01]

Trimming can both be good and bad. It would probably be a good idea to trim off some really low quality bases (ie <10). If nothing else it will make things computationally easier. Generally, the trade off between more sequence and higher quality sequences evens out in terms of assembly quality. It just means with more sequences you'll need more RAM and more CPU time to get the job done.

However, you should think about how you do your assembly some. Longer kmer assemblies will require higher quality data because the chance of unique kmers due to sequencing errors increase with greater values of k. So, you could try lower values of k with lower quality score trimming, and higher values of k with more quality score trimming. But be mindful of how much sequence you're losing.

One thing you can do is inspect your kmer coverage distribution with different combinations of k and quality cut offs. You can do this fairly quickly using Jellyfish and make plots of the resulting distribution file it outputs. The basic point is you want something with nice big peak coverage out around 20x or greater.