RNA seq: handling samples after re-sequencing

[Siva]

I set up an RNA seq experiment with 4 samples and 4 bioreplicates each. I now have a total of 16 fastq files. However, of these, 8 are going to be resequenced because number of reads were too low (though still usable). I was told that it was technical issue related to quantifying the library at the sequencing facility. When I get the new set of 8 files, how should I handle them:

1. As a completely separate technical replicate? (The same library will be used for re-sequencing)
2. Merge the BAM files together? (e.g., BR1a.BAM + BR1b.BAM)....concatenate BAMs or merge the raw reads together? (e.g.., BR1a.fastq + BR1b.fastq) concatenate fastqs?

Which of the above will give me greater statistical power. How will I make better use of the fact that I will have resequenced the same library deeper?

Thanks
Siva

[Bukowski]

It depends what your intention for the experiment is surely? If you need extra depth you could just combine the Fastq and align.

If you're just interested in differential expression then this paper might convince you to use them as biological replicates:

http://arxiv.org/abs/1301.5277

[Siva]

It depends what your intention for the experiment is surely? If you need extra depth you could just combine the Fastq and align.

If you're just interested in differential expression then this paper might convince you to use them as biological replicates:

http://arxiv.org/abs/1301.5277

My immediate aim to find differentially expressed genes. However I am also interested in more depth. Will be useful in analyzing novel genes.

I will go through this paper. However, I was under the impression that treating them as a bioreps might actually lead to higher occurrence of type 1 error in differential expression analysis.

Thanks much
Siva

[chadn737]

If they are from the same libraries then they are not biological replicates and should not be treated as such.

In cases where you multiplex samples and spread them across different flow cells or runs, you can simply combine them as lane effects will apply to all the samples. To see a detailed description of this, read the paper by Auer and Doerge: http://www.genetics.org/content/185/2/405.abstract

However, since you will only be doing further sequencing on 8 of the samples, this is a far more complicated design. I would say if you want to keep all the data, then you should use the multifactorial capabilities of tools like DESeq or EdgeR to account for this.