We collected targeted metabolomics (concentrations in uM) from human plasma samples (n=8 (crossover)). It is unclear if internal standard normalization is enough for data analysis or if I should also normalize date using the median or PQN etc.
They are designed for different purposes:
- The internal standards (IS) are mainly used for identification and quantitifcation of individual compounds - the hallmark of targeted metabolomics;
- Statistical normalization aims to address the global distribution of data to make features more comparable across samples - this is independent of your internal standards
In your case, sample-wise row normalizations (like PQN or Median) are designed to correct for variable biological sample dilution—such as in urine, where hydration levels fluctuate wildly between participants. In plasma assays, the sample volume input is strictly controlled during sample prep. Some general comments:
- Skip PQN / Median Normalization entirely. Your fixed plasma input volume combined with IS-based absolute quantification already resolved sample concentration scaling.
- Apply Log Transformation Biological concentration data is typically right-skewed and heteroscedastic (variance increases as concentration increases). Log transforming stabilizes this variance.
- Use Crossover-Appropriate Statistics. Since this is a crossover study, samples are paired. Perform paired statistical modeling (e.g., paired t-tests or linearr models) to account for subject ID and period effects) on your log-transformed values.