454 and homopolymers

[adaptivegenome]

I am planning to use CORTEX to assemble 454 data and I want to use the cut-homopolymer option but I am having trouble deciding at what size homopolymers are a problem for 454. Seems to me like it would somewhere in the 4-6 base range, however I would like to disrupt my reads as little as possible.

Does anyone has some insight into this? Has anyone looked at error rates as a function of homopolymer length in 454?

Thanks,
David

[Zam]

Hi there

I wrote Cortex, so I know about that, but I have somewhat limited 454 experience.
If you want to be systematic, you can

1. load in your reads multiple times, and each time use a different homopolymer threshold, and use the --dump_filtered_readlen_distribution option to dump a file showing how it affects your read lengths. That tells you how much read-length you are throwing away

2. Theres the issue of 454 homopolymer errors, and at what length they are prevalent. I can't help with that to be honest.

If you are just making variant calls, as I was when I did this, I would use a limit of 3 and see how I go, but that's quite conservative.

[adaptivegenome]

Hi there

I wrote Cortex, so I know about that, but I have somewhat limited 454 experience.
If you want to be systematic, you can

1. load in your reads multiple times, and each time use a different homopolymer threshold, and use the --dump_filtered_readlen_distribution option to dump a file showing how it affects your read lengths. That tells you how much read-length you are throwing away

2. Theres the issue of 454 homopolymer errors, and at what length they are prevalent. I can't help with that to be honest.

If you are just making variant calls, as I was when I did this, I would use a limit of 3 and see how I go, but that's quite conservative.

I wrote a script to do #1 and if you set the threshold to 3 or 4, it seems like a lot of reads get fragmented beyond use. I was hoping to balance the threshold with the desire to keep as many reads as possible. It is hard to do this, however, without knowing at what length the error rate explodes for homopolymers. I suppose I could determine this but I am hoping someone on seqanswers has some experience here.

[Zam]

Fair enough - good luck!