Data normalization is intended to reduce technical noise, and this strongly impacts the variance from technical and/ or experimental parameters in RNA seq. We have seen parallel libraries produced from the same RNA contribute more to variance than the genetic background of the individual organism.  The secret is to replicate, replicate, replicate if you really want to be able to draw strong conclusions from the data because there are so many procedural steps in RNA-seq where variance can be introduced.

Dear Jan,

Whenever we look at results from an RNAseq Data two very important parameters, should be constant if not suggested otherwise by the experimental design.

One is the time of collection of sample from a particular tissue type even in an in vitro condition as the variation in transcripts may be as a result of the variation of expression of different genes at different stages of the cell cycle.

The second is the platform of sequencing which I presume you are keeping constant.

Hence if the above parameters are constant then I believe that you can try making biological replicates of the different samples to create a map of expression values for each sample and then compare across samples.

Hope this helps

Cheers

The variation could be a mixture of both experimental and biological origin. To overcome the experimental variation, technical replicates (measuring the same sample multiple times) are helpful.  Biological variations are inevitable, and the magnitude will be relatively higher in tumor specimens (heterogeneous cell population) compared to normal tissue.

Variation across the samples collected at different times should not be unexpected. In fact, such variation will have hidden biological meaning.

The proposed speculations are subjective to sample type.

Example: in case of normal tissue, the chances of homologous transcripts expression (across the sample) could be very less. But, this can be a major component in case of tumor specimens.

In addition to the biological variance (sample collection, time points, genetic background, etc), there is a lot of variance introduced by some experimental parameters, especially library preparation. RNA isolation can also play a role. A strategy to overcome this is pooling the biological samples and creating multiple libraries, this should give you tighter replication and hopefully make it easier to identify differential expression between your experimental samples.

Well, it's not very clear from your question if you're looking at samples from the same tissue of one individual, the same tissue over several individuals (which species?), or the same tissue type... If you want the maximal reproducibility over different samples, you should, if possible, sample your tissues always at the same time of the day (to avoid variability due to circadian rhythms, often overlooked), extract your RNAs, prepare your libraries and sequence them at the same time with the same methods (to avoid technical variability). Then variability is also depending on species: it's greater in humans, even if you try to take samples from homogeneous populations (same gender, age, genetic background...) than in congenic mouse strains... Hope that helps!

Thanks for the answers, but this is not what I had been looking for. Let me reiterate.  

I'm not expecting to see the same expression, but I was curious about the causes of the variability I see. 

I don't have access to more samples so I can't make inferences of my own. Based on your experience, is the majority of the variability random experimental error or is it inherent to the host the sample had been collected from?

Another way of asking this question would be: how much of the variation can be explained by eQTLs and other epigenetic markers?

Regarding your example of Heisenberg's uncertainty - this is exactly what I'm asking about! However, I don't think you understand it well. 

The uncertainty principle is a fundamental and inherent property of quantum systems and it is not equivalent to the "observer effect", which you are talking about. 

thanks!

Data normalization is intended to reduce technical noise, and this strongly impacts the variance from technical and/ or experimental parameters in RNA seq. We have seen parallel libraries produced from the same RNA contribute more to variance than the genetic background of the individual organism.  The secret is to replicate, replicate, replicate if you really want to be able to draw strong conclusions from the data because there are so many procedural steps in RNA-seq where variance can be introduced.