What are the possible dangers of using inside sample and between samples normalization exchangeably?

Question

[jack]

Normalization is one of the main preprocessing step in bioinformatics data analysis. most of the time people just talk about the normalization methods that they have applied to their own data and they don't talk about that, whether they did nside sample and between samples normalization .

For me it's not clear that, when I should use inside sample normalization and when between samples normalization? Is there any rules of thumb for that?

[Devon Ryan]

Could you rephrase your question? It doesn't parse correctly into English as written. Also, the goal of the analysis can be important when thinking about normalization, so mention that as well.

[jack]

Devon, I edited it.

Answer 1

[Istvan Albert]

The dangers of any type of normalization is that we alter the real, meaningful changes that are present in the data.

It is difficult to predict beforehand which type of normalization is more suited to any give dataset as the data preparation, acquisition and experimental design may all have an effect. In general most normalization methods rely on the expectation that for each sample the distribution of expression levels is the same for all conditions and that most genes are not differentially expressed across conditions.

To answer your question I would say that

• within sample normalizations will increase the rate of false negatives as any error or outlier will affect the consistency of the group
• across sample normalization we alter data based on prior expectations (groups) where outliers and errors make the group more different and thus make it more likely to detect significant differences between conditions. We run the risk of raising the rate of false positives.