[lhlbio]

I was so inspired by your ideas
"My take is that if you are so deep into a process that you have lost track of what it is you are ultimately trying to achieve, you need to step back and re-think what it is you are doing and why you are doing it."

Thank you so much!
--
huilili

[pmiguel]

Hi Y'all,

this thread seems pretty appropriate for my question:

I would like to compare rates of synonymous and nonsynonymous substitutions among sequences of orthologous genes from ~ 8 species of fish in the family Embiotocidae (closest related genome sequenced would be one of the cichlids) to test for positive Darwinian selection (genome size is ~ 1 GB). I am relatively poor and would love to "goldilocks" it on my first run. That is, how many fishes can I multiplex into a single perfect firmness bed, err I mean Illumina highseq lane and still get an acceptable amount of coverage to call polymorphisms between orthologs.

thanks very much!

Goldilocks tried all 3 beds before finding one of a firmness she was satisfied. And your situation is insanely more complex than Goldilock's.

Questions of this sort are nearly impossible to answer with any accuracy. You do understand that any organism/tissue will have vastly variable ranges of expression diversity? Best case, you only need 10 genes for your study and those 10 genes happen to be 90% of the expression in the organism/tissue you happen to choose. I won't go worse case, because that would involve something like you falling into the tank with your fish and being eaten alive.

Can you get by with 6 samples (including reps)? If so try that in one lane and see how it goes. Oh, don't discount the informatics chops it takes to process 1 lane of Illumina sequence (30-50 billion bases of sequence if you chose the 2x100 PE option).

Shouldn't you do something like RADseq for an experiment of this nature rather than RNAseq?

--
Phillip

[sdriscoll]

Unfortunately calling SNPs requires quite a bit more coverage than the typical gene expression experiment. The reason being that you'll want to put some type of minimum count level requirement on the candidate sites similar to how you would put a min read count limit on genes in a differential expression test. In a DE test we might cut out genes with less than 10 hits or maybe even those that have less than 1 hit per million mapped reads (edgeR advice). If you consider potential aligner error there's possibly a 5 to 10% chance any read is incorrectly aligned (depending on your aligner) and with the high variance of low count levels you're going to want to apply that minimum of 10 hits at every candidate site. Of course there's sequencing error on top of that and god knows how much other noise in the data. Maybe you can see where I'm going with this - but for gene expression which is typically dealing with features 1000's of kb in length we might accept 10 hits as a cutoff which is clearly well below having a lot of coverage of a gene. In my experience 20M reads typically yields ~14,000 testable genes in the mouse genome at such a cutoff.

For SNPs you want that kind of depth at the single base resolution. So what's going to happen, if you only have 20 million reads or so, is you're going to limit your results to a small percent of expressed genes. That's not to say you won't have some results - but they may be limited to higher expressed features. Since nobody can prove that low expressed genes don't matter that's gotta make you at least a little uncomfortable.

[Glongo]

Hi Y'all,

this thread seems pretty appropriate for my question:

I would like to compare rates of synonymous and nonsynonymous substitutions among sequences of orthologous genes from ~ 8 species of fish in the family Embiotocidae (closest related genome sequenced would be one of the cichlids) to test for positive Darwinian selection (genome size is ~ 1 GB). I am relatively poor and would love to "goldilocks" it on my first run. That is, how many fishes can I multiplex into a single perfect firmness bed, err I mean Illumina highseq lane and still get an acceptable amount of coverage to call polymorphisms between orthologs.

thanks very much!

[chadn737]

And what do you plan to do with these other samples? Have you guys given any thought to how you will analyze the data? Will attempt to construct a transcriptome denovo? Or are these other species closely related to Tomato or another sequenced genome where you could map to that and obtain reasonable data?

It doesn't make sense to blindly sequence without thinking the experiment through and knowing what you need up front. Its also very hard for people to give advice when you are very vague about the details of your experiment and end goals.

[vebaev]

yes one of the plants is tomato but other samples we have plants with no ref genome/annotation

[chadn737]

no, each 2 are different plant, 1 control and 1 treatment , no info of transcriptome size (probably only one of them will be sequenced with ref genome - tomato)

1) If you want to find differentially expressed genes....you need more than 1 replicate per condition.

2) If the reference genome is Tomato, which has had its genome sequenced and at least a version one annotation, then you should be able to estimate the transcriptome size. Unless I have misunderstood and this is not Tomato.

[vebaev]

no, each 2 are different plant, 1 control and 1 treatment , no info of transcriptome size (probably only one of them will be sequenced with ref genome - tomato)

[chadn737]

And if we came back on the topic what is the optimal number of millions reads for de novo RNA-seq without reference genome? as if we have 12 samples per 6 lanes = ~15M for sample is this OK for this task?

What do you want to do with the data? Are the 12 samples from the same plant? Do you have a close relative sequenced? Do you have any rough estimation of the genome and transcriptome size? You need to be clear about what exactly it is you want to do, there is no simple answer to these questions.

Even for Arabidopsis I would want more than 15M per sample.

[vebaev]

And if we came back on the topic what is the optimal number of millions reads for de novo RNA-seq without reference genome? as if we have 12 samples per 6 lanes = ~15M for sample is this OK for this task?

[chadn737]

I am not sure I understand the benefit of multiplexing in this case. What chance are you referring to?

If there are lane to lane effects....which there usually are to some extent, then multiplexing all your samples means that those lane effects will affect all samples, not just one. It may be minor, but its a matter of good experimental design to reduce the possible technical variables as much as possible. Statistically this is the optimal experimental design.

http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2881125/

Let me reverse the question. What do you gain by NOT multiplexing?

[GenoMax]

I see no reason not to multiplex. Lane effects may or may not be an issue, but why take the chance when it is so easy to multiplex.

I am not sure I understand the benefit of multiplexing in this case. What chance are you referring to?

[chadn737]

If there are only 2 samples is there any advantage in multiplexing them (unless you are worried about lane effect of some kind)?

You could just do 3 lanes of each sample. Since you do not know the size of the transcriptome you would want to get as many reads as you can.

I see no reason not to multiplex. Lane effects may or may not be an issue, but why take the chance when it is so easy to multiplex.

[GenoMax]

If there are only 2 samples is there any advantage in multiplexing them (unless you are worried about lane effect of some kind)?

You could just do 3 lanes of each sample. Since you do not know the size of the transcriptome you would want to get as many reads as you can.

Thanks.
The goal is to compare 2 samples of RNA-seq of unsequenced plant, so we do not know the size of the transcriptome. What should be the optimal M reads for this de novo RNA-seq task?
I asked since we will have 6 lanes but other colleagues will want to put also their samples (similar as ours samples) , and wanted to decide how much samples in total we can multiplex and load on these 6 lanes to be optimal for this task?

[vebaev]

Thanks.
The goal is to compare 2 samples of RNA-seq of unsequenced plant, so we do not know the size of the transcriptome. What should be the optimal M reads for this de novo RNA-seq task?
I asked since we will have 6 lanes but other colleagues will want to put also their samples (similar as ours samples) , and wanted to decide how much samples in total we can multiplex and load on these 6 lanes to be optimal for this task?