1
$\begingroup$

I have data from a lipidomic experiment comprising: 2 conditions and 5 samples in each condition; and for each lipid the value is a peak area. I am doing a lipidomics analysis using the web app lipid ontology, and they recommend to normalize the data prior to doing the analysis,

"There are many ways to normalize data, so LION/web cannot check whether data is properly normalized. If you don't know how to normalize the data, we recommend normalization by expressing all lipid species as fraction of the sum. A popular website that supports several normalization algorithms is MetaboAnalyst."

I don't know how to normalize my data, I cant find any paper related to lipidomics normalization to know to best way to do that, and also I can't find normalization function in the MetaboAnalyst.

Thank you in advance!

$\endgroup$
4
  • $\begingroup$ Hi @Mee have you run controls common to all data sets? Are you sure this is not 'standardisation' rather than 'normalisation'? $\endgroup$
    – M__
    Oct 28 '20 at 16:22
  • $\begingroup$ Hi @Michael, they called Normalization all the time in the lipid ontology website. What would be the difference? Thanks $\endgroup$
    – Mee
    Oct 28 '20 at 17:37
  • $\begingroup$ Hmmmm I'll make a when I have a moment. There is a very real difference between normalisation and standardisation from a statistics point of view. In my view this is standardisation. I need to know details about your runs. Were all runs simultaneous? same day, same time, same apparatus? $\endgroup$
    – M__
    Oct 29 '20 at 1:41
  • $\begingroup$ ok, I know what you mean with normalization and standardization. Still, they called it normalization in the website... The samples were not run simultaneously, but they were run in the same conditions.. $\endgroup$
    – Mee
    Oct 29 '20 at 8:40
1
$\begingroup$

Basically and briefly,

  • Normalisation = transforming a data set to conform to a normal distribution. Converting to a frequency is NEVER a normalised transformation
  • Standardisation = ensuring each point of the data set is comparable to the others, either directly or relatively depending on the calculation

The information you have received is incorrect, moreover you only convert to a frequency for a final assessment of the data and in my experience NEVER as an intermediate step. Normally data point between 0 and 1 are bad news and often require a transformation to get them away from fractions.

I don't understand the behaviour of the variance involved in metabolism and particularly lipids, however you should refer this issue to "cross-validation" Stackexchange. I did have some understanding of enzyme kinetics but in a different area.

Cross-Validation Stackexchance statisticians will understand less of the biology than I do (which is a bit worrying of course), so you will need to give them quite detailed information on the data set. They will almost certain support my concern of the advice you are given and provide a better method of standardisation. One method is to take the difference in mean and divide it by the standard deviation, you can then normalise the standardised data. They will want to know the goals of your investigation to understand who to process the data further.

What I can say is variation between runs is often an issue (in your apparatus perhaps not, I dunno) and standardising against a common control is one approach to overcome this, e.g. the control same common to all runs would be the mean about which all other data is centred.

What I suspect is that lipids are less popular than other fields, so standard protocol may not be observed.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.