Correcting a reference genome for Tophat

[tirohia]

Hi.

I am currently mapping plant transcriptome reads from one cultivar to the genome of another culitvar of the same species. It has been suggested that to improve the mapping, it might be worthwhile to correct the reference genome for use with cultivar that my reads are from.

I can see how this could improve the number of reads that map to the reference. It feels like it might be a bit circular though, as if I'm taking the transcriptome from one of my samples and using that as reference for itself (and the rest of my samples). Can anyone think of why this might be bad idea?

At the moment I am using tophat and taking the resulting insertions.bed and deletions.bed, reading the files manually and then manually ineserting these into the reference genome. Is there a better way to do this?

Cheers
Ben.

[Brian Bushnell]

Hi.

I am currently mapping plant transcriptome reads from one cultivar to the genome of another culitvar of the same species. It has been suggested that to improve the mapping, it might be worthwhile to correct the reference genome for use with cultivar that my reads are from.

I can see how this could improve the number of reads that map to the reference. It feels like it might be a bit circular though, as if I'm taking the transcriptome from one of my samples and using that as reference for itself (and the rest of my samples). Can anyone think of why this might be bad idea?

At the moment I am using tophat and taking the resulting insertions.bed and deletions.bed, reading the files manually and then manually ineserting these into the reference genome. Is there a better way to do this?

Cheers
Ben.

If you want to improve mapping without bias, you need a new mapper, or a new assembly. So, either get a better mapper, or assemble reads from your sample denovo (which is very difficult), or map whole genome reads to the reference, then correct the reference by replacing the reference with any homozygous variation calls. This last method is the best as long as you have no reference bias. Which you will. But still, it could be the best if the ref-bias is minimal.

Mapping to transcriptomes is already highly subjective, biased, and error-prone - no matter how robust the rest of your study may be. But mapping to a modified transcriptome based on RNA-seq data from a different strain should never get through peer-review.

Therefore... you need whole-genome data. Exome, transcriptome, or whatever limited subset are not alone mathematically adequate in describing a genome - and thus, they cannot be used definitively when analyzing a related strain. If you just modify a transcriptome arbitrarily based on your reads, then remap, you will get indefensible data.

To reduce bias to minimum, when cross-strain mapping, you should use the most sensitive and accurate mapper available. Using iterative stages of "map-changeReference-map-changeReference" will exponentially amplify bias.

-Brian