Hi Juliano,

I think PQN would be a good starting point.

Dieterle F, Ross A, Schlotterbeck G, Senn H. Probabilistic quotient normalization as robust method to account for dillution of complex biological mixtures. Application to 1H NMR metabolomics. Analytical Chemistry. 2006;78:4281–4290. doi: 10.1021/ac051632c. PMID: 16808434

If you use Matlab I can send you the code.

I also suggest you read:

Veselkov, K. A., et al, Optimized Preprocessing of Ultra-Performance Liquid Chromatography/Mass Spectrometry Urinary Metabolic Profiles for Improved Information Recovery. Analytical Chemistry, 2011, 83(15): 5864–5872.

and

State-of-the art data normalization methods improve NMR-based metabolomic analysis.Kohl SM, Klein MS, Hochrein J, Oefner PJ, Spang R, Gronwald W. Metabolomics. 2012 Jun;8(Suppl 1):146-160. Epub 2011 Aug 12. PMID: 22593726

You may also want to have a quick look at Metaboanlayst (online tool): www.metaboanalyst.ca .It has a few normalization methods that you may which to quickly try out.

Xia J, Mandal, R, Sinelnikov IV, Broadhurst D, Wishart DS. MetaboAnalyst 2.0-a comprehensive server for metabolomic data analysis. Nucleic Acids Res. 40(Web Server issue), W127-133 (2012).

PQN has been very popular with NMR data which is well known to be very reproducible (w.r.t. instrument precision), however, I personally have not seen a massive improvement on LC-MS data with this method (most of my work is serum though) However, there are more complex issues that arise with LC-MS data - regarding instrument precision over time. Columns degrade and MS instruments get dirty. This results in signal attenuation which has nothing to do with sample volume (but will effect the ability for PQN to work properly). I am a big proponent of using Quality Control samples to monitor "drift" and then correct for the drift using software.

I have attached a recent review:

Dunn WB , Wilson ID, Nicholls AW Broadhurst D. The importance of experimental design and QC samples in large-scale and MS-driven untargeted metabolomic studies of humans Bioanalysis September 2012, Vol. 4, No. 18, Pages 2249-2264 , DOI 10.4155/bio.12.204

I hope that helps.

VBW, David.

These two papers could help you:

Dieterle et al. Probabilistic Quotient Normalization as robust method to account for dilution of complex biological mixtures. Application in 1H NMR metabonomics. Anal. Chem. 2006, 78, 4281-4290

R.J.O. Torgrip et al. A note on normalization of biofluid 1D 1H NMR data. Metabolomics 2008, 4:114-121

Both use the full spectrum to evaluate the dilution factor (instead of creatine), however they were develloped for NMR data, this will require adaptation for the HPLC-MS data.

Thank you very much Lionel.

I'm also in favor of using Probabilistic Quotient Normalization for urine. In fact you can find a very useful paper which focused on that particular question:

Veselkov, K. A., et al, Optimized Preprocessing of Ultra-Performance Liquid Chromatography/Mass Spectrometry Urinary Metabolic Profiles for Improved Information Recovery. Analytical Chemistry, 2011, 83(15): 5864–5872.

Hope this helps!

i've used statistical data normalization with CODELINK package for R software, by the median normalization method

You may try to normalize by TIC (total ion chromatogram). We also work on urine metabolomics and in our case it worked for SRM analysis. Good luck!

I've also had same experience that normalization doesn't work well. Sometime, without normalization makes better F1 score than with normalization methods including linear regression method, quantile, group-based approach to find artificial metabolite biomarker using FTMS data.

Hi Juliano,

I think PQN would be a good starting point.

Dieterle F, Ross A, Schlotterbeck G, Senn H. Probabilistic quotient normalization as robust method to account for dillution of complex biological mixtures. Application to 1H NMR metabolomics. Analytical Chemistry. 2006;78:4281–4290. doi: 10.1021/ac051632c. PMID: 16808434

If you use Matlab I can send you the code.

I also suggest you read:

Veselkov, K. A., et al, Optimized Preprocessing of Ultra-Performance Liquid Chromatography/Mass Spectrometry Urinary Metabolic Profiles for Improved Information Recovery. Analytical Chemistry, 2011, 83(15): 5864–5872.

and

State-of-the art data normalization methods improve NMR-based metabolomic analysis.Kohl SM, Klein MS, Hochrein J, Oefner PJ, Spang R, Gronwald W. Metabolomics. 2012 Jun;8(Suppl 1):146-160. Epub 2011 Aug 12. PMID: 22593726

You may also want to have a quick look at Metaboanlayst (online tool): www.metaboanalyst.ca .It has a few normalization methods that you may which to quickly try out.

Xia J, Mandal, R, Sinelnikov IV, Broadhurst D, Wishart DS. MetaboAnalyst 2.0-a comprehensive server for metabolomic data analysis. Nucleic Acids Res. 40(Web Server issue), W127-133 (2012).

PQN has been very popular with NMR data which is well known to be very reproducible (w.r.t. instrument precision), however, I personally have not seen a massive improvement on LC-MS data with this method (most of my work is serum though) However, there are more complex issues that arise with LC-MS data - regarding instrument precision over time. Columns degrade and MS instruments get dirty. This results in signal attenuation which has nothing to do with sample volume (but will effect the ability for PQN to work properly). I am a big proponent of using Quality Control samples to monitor "drift" and then correct for the drift using software.

I have attached a recent review:

Dunn WB , Wilson ID, Nicholls AW Broadhurst D. The importance of experimental design and QC samples in large-scale and MS-driven untargeted metabolomic studies of humans Bioanalysis September 2012, Vol. 4, No. 18, Pages 2249-2264 , DOI 10.4155/bio.12.204

I hope that helps.

VBW, David.

Juliano,

When measuring urines for infectious events you should not use nroamilzation against creatinine. The formation rates of creatine andhence creatinine are altered by changes in the urea cycle function which is highly manipulated by the host during various stages of the jnfectious event, nitric oxide production rates, levels of glycine, upregulated of arginase etc. The best way top ensure you have a proper reading of the results is to do both quantitative and relative abundances. both of these anyalses give you quite different results. the raw or absolute data will give you data on kidney function and the factors such as diuresis events etc, On the other hand the relative abundance gives you data on the relationships between the m,etabolites and this give valuable information on the underlying metabolic pathway regulations and activity. normalizing to creatinine tedns to mask many of these features. what I would do though to proove it to yourself is run all three analyses and you will see what we found when we first looked at the urine changes.

Neil McG

Further to my last reponse. If you gow the Leishmania in minimal media (or at least one to sustains its viability and then assess the metabolites release. We did this with E.coli and found that the pattern of metabolite in the urine matched tha pattern the microbe produced. The levels of the metabolites varied in relatinship to the colony count on the organism in the bowel. Your microorganism will be there at lower concentrations and may not be detectable by this method but it is worth an assessment.

Hallo colleagues

There are a lot of new nice normalisation methods for me, thanks for that!

So but there is allways one fundamentel problem behind, if you try to normalise metabolomic data - normalisation only works if your samples are quite similar in the response of the metabolite profile! If you have strong differences in your measured response, it is nearly impossible, because of the heterogeneous chemical character of your metabolite, some are well corrected with one method some not..

We hat urine from control mouse against diabetic mouse and obese mouse, even the control group was quite difficult to homogenise.. then the diabetic mouse had such enormous sugar content, which eleminate the concentrations of all other metabolites, so we were never sure that we calculated well..

However normalisation against osmolarity worked quite ok but not perfect!!

So what I want to say is, there is no good method available at the moment, mayby never.

Good luck Juliano

I'd like to echo Christian in saying that normalisation to total intensity is a bit flawed as it can vary a lot, and this method tends to put too much weight on high intensity peaks - so what if they're the ones changing? Can end up with results which are not strictly true. It's probably a bit late, but the best way to normalise urine is to normalise it to volume per time period (e.g. mL/24 h). This is normally what creatinine is an estimate of, though as pointed out earlier, under certain conditions such as yours, it's possible not the best marker of dilution. So if you can collect all urine over a 12 or 24 h period, you don't need to worry about things like total intensity (effect of ion suppression over the course of a run?), and you can express your results per 24 h, which is translatable to targeted measures. So it's easier to compare your results with past literature.

Dear Dr. Broadhurst,

Can you share the PQN matlab code please ?

Thanks in advance,

Guillaume

Hi Juliano,

normalization in metabolomics data analysis is a quite complex subject and a golden standard does not seem to be established yet. We investigated the performance of different normalization methods, including PQN and  total spectral intensity, and could report large differences in the detected biomarker signatures. Therefore, we recommend, instead of using one of the suggested normalization strategies, to choose statistical analysis tools, which are insensitive to a chosen scaling or normalization method, namely the newly developed zero-sum (logistic) regression. It enables the detection of metabolic/clinical biomarkers independently of the chosen normalization method (actually, no data normalization is needed at all). We believe that this approach leads to unbiased molecular signatures, which can be easily transferred between different labs and easily from bench to bedside. Have a look at the attached publications.  

Article Reference point insensitive molecular data analysis

Article Scale-Invariant Biomarker Discovery in Urine and Plasma Meta...

Enter your answer

Normalization of human urine samples is not trivial task,

I would like to add that in any experimental design, samples selection and collection are very important and crucial steps, For example fasting is essential and the time of sample collection is also crucial.

We discussed all factors that could affect the samples quality in a recent published paper entitle

(Standardizing the experimental conditions for using urine in NMR-based metabolomic studies with a particular focus on diagnostic studies: a review)  

https://link.springer.com/article/10.1007/s11306-014-0746-7

I hope this paper will give you some insight about all factors that may affect the urinary analysis