Hi 454 analysers,

We are doing resequencing experiments here and are developing our own BLAT and a DB based mapping and SNP discovery pipeline.

Finally we are able to detect SNVs, Isertions, Deletions and InDels. We have already validated the pipeline with random errors and known errors. But now the final part appears to be more tricky. The 1 million dollar question: is it a heterozygous or a homozygous variation.

For example: 400x and 25% or 80% error rate, what do you do with that? In very long stretches of homopolymers, for example 10Cs. Chances are very big that you get 20% fake error of 1 or 2C extra.

Well we are making a mathematical model to determine the cutoff frequencies of error rates at a certain coverage. The higher the coverage is the narrower the band becomes wherein a heterozygous error rate can be, but how narrow?

I know the average error rate of 454 is around 1/1000, but that is not what i need, because Single nuc variations get filtered out in a very early stage (i filter everything that is <20%). The residual errors are either true variations or very frequent errors (such as homopolymers possibly?).

Ok what i need is some kind of homopolymer error rate. I suppose it is linked to the length of the homopolymer, the longer it is, the more probable it is that random errors will occur. Is there a function known that gives the error rate ~ homopol length? I can calculate it myself, but i have some gut feeling this might not be this easy.

Is anyone aware of a good article that describes different error rates of different types of errors? I have the article Pyrobayes: an improved base caller for SNP discovery in pyrosequences, but that only describes general error rates for substitution, deletion and insertion, not for homopolymer or normal.

In the near future we will start with bisulfite treated amplicon sequencing, and with only 3 nucs, there will be even a bigger homopol error rate, and i would like to investigate/model some freqs of certain things upfront so that i can determine a useful coverage.

We are doing resequencing experiments here and are developing our own BLAT and a DB based mapping and SNP discovery pipeline.

Finally we are able to detect SNVs, Isertions, Deletions and InDels. We have already validated the pipeline with random errors and known errors. But now the final part appears to be more tricky. The 1 million dollar question: is it a heterozygous or a homozygous variation.

For example: 400x and 25% or 80% error rate, what do you do with that? In very long stretches of homopolymers, for example 10Cs. Chances are very big that you get 20% fake error of 1 or 2C extra.

Well we are making a mathematical model to determine the cutoff frequencies of error rates at a certain coverage. The higher the coverage is the narrower the band becomes wherein a heterozygous error rate can be, but how narrow?

I know the average error rate of 454 is around 1/1000, but that is not what i need, because Single nuc variations get filtered out in a very early stage (i filter everything that is <20%). The residual errors are either true variations or very frequent errors (such as homopolymers possibly?).

Ok what i need is some kind of homopolymer error rate. I suppose it is linked to the length of the homopolymer, the longer it is, the more probable it is that random errors will occur. Is there a function known that gives the error rate ~ homopol length? I can calculate it myself, but i have some gut feeling this might not be this easy.

Is anyone aware of a good article that describes different error rates of different types of errors? I have the article Pyrobayes: an improved base caller for SNP discovery in pyrosequences, but that only describes general error rates for substitution, deletion and insertion, not for homopolymer or normal.

In the near future we will start with bisulfite treated amplicon sequencing, and with only 3 nucs, there will be even a bigger homopol error rate, and i would like to investigate/model some freqs of certain things upfront so that i can determine a useful coverage.

## Comment