Qustion about miRNA-Seq normalization and analysis

[yiwenhe]

Hi, I am working on a set of miRNA TruSeq data. I did some search here and have decided to try TMM and DESeq for normaliztion. There are 3 biological repeats for each condition, but when the core facility ran the experiment, each sample was split into multiple lanes/flow cells. My question is what should I do with these technical repeats. My intention is to sum those up so that I have one value per gene for each biological sample, and normalize on that. But I don't know how it is different from normalization across all the technical/biological repeats.

Thank you for any input.

[dpryan]

Summing the technical replicates is the correct solution in RNAseq.

[yiwenhe]

Thank you!

[nsmehta]

I have been trying to find the answer to this, but why do you need to sum the lanes? I have 4 groups with 4 biological replicates per group and 3 technical replicates for each bio rep. I have found some differences in frequency distribution between technical replicates, but not consistently on one lane or another. Is there a way to handle this?

[gringer]

I have found some differences in frequency distribution between technical replicates, but not consistently on one lane or another.

It has been previously found that variation in technical replicates is not significant for the purpose of calculating differential expression (Simon Anders discusses this in his DESeq paper, referencing Marioni et al. and Bullard et al.). From the DESeq documentation:

The count values must be raw counts of sequencing reads. This is important for DESeq's statistical model to hold, as only the actual counts allow assessing the measurement precision correctly. Hence, please do do not supply other quantities, such as (rounded) normalized counts, or counts of covered base pairs -- this will only lead to nonsensical results.

Furthermore, it is important that each column stems from an independent biological replicate. For technical replicates (e. g. when the same library preparation was distributed over multiple lanes of the sequencer), please sum up their counts to get a single column, corresponding to a unique biological replicate. This is needed in order to allow DESeq to estimate variability in the experiment correctly.

You need to sum the lanes (at least for DESeq) because the algorithms used assume a particular model, namely that each count represents a different biological sample.

As a general rule, if you choose to break the rules of a particular statistical test, you need to make people aware that you're doing that, and also mention your [good] reason for doing that.