Murph's metabolomics analysis

[Murph]

I've been looking at the outputs of various metabolomic papers. I'm interested to see where they agree and disagree. Is there much we can glean from them or is there just a lot of noise?

In this thread I intend to post some of the outputs of this investigation.

For starters, a nice simple one. How similar are the measurements of amino acids in serum between the 2012 Armstrong metabolomic paper and the 2021 Norwegian paper by Hoel, Fluge, Tronstad et al? The answer is: mostly similar, but with outliers on hypoxanthine and glutamate.

I have data from a bunch more studies, including Naviaux's big 2017 study, a lipkin study, Hanson's 2018 study and her recent study where she took measurements four times before and after two bouts of exercise.

My eventual goal is to create a website that permits people to make charts like the above one themselves. It would let us investigate the history of any chemical or metabolite that is mentioned by researchers, see if it has been measured consistently high or low, is inconsistent, or has not been measured before.

Hopefully by gathering all the information together we can get more clarity on what's consistently out of line. Another possibility is collating all the information lets us conclude that one study disagrees with all the rest and is, perhaps, low quality. Or alternatively perhaps there will be so many outliers and disagreements in the mix that we have to conclude that metabolomics is not a reliable method or source of information for me/cfs cohorts as defined.

 

[Murph]

The below is an output from my analysis of the Naviaux data. A strength of his data is it is reported per patient, not just as an average.

What I draw from this chart is the idea that hypoxanthine levels can be vary variable. Some people are big outliers. It might explain why this molecule is an outlier on the chart above - one or two such outliers included in the controls vs the patients could explain the discrepancy.

Notably the medians are probably higher in the patients, even though the average is higher for the controls. This reminds me: it's probably a good idea to let people measure medians and means separately when I get round to giving this a public interface.

For now, my analysis of naviaux is concerning though. I can't get his measurements of ratios in cases and controls to agree with the Norwegians at all. I need to do some more checking - it's possible the error is mine.

 

[Murph]

Here's another one. Lots of detail here. It compares Hanson's 2018 metabolite paper to Fluge's 2021 paper. I've added lines of best fit, and used log scales on the axes here. You can see there's an upward slope to the lines of best fit in most categories, which means the papers agree generally. Obviously though the match is not 1:1.

 

[marcjf]

Would comparing be hard for some papers? The one from Fluge 2021 shows multiple subset of patient profiles. Given that ME/CFS is not an actual disease, but a syndrome, I think we are bound to see profiles that do not match, and just translate to similar symptoms. Summarizing them could be misleading.

I think it would be more useful if we looked at our own metabolite profile, and then see if they matched any of these papers. I haven't seen anyone here doing that yet. When looking at mine for example, I can see some things matching what Hanson 2018 found, like the disturbed Taurine pathway (I have borderline low levels, despite not being a vegan). On the other hand, I found exactly the opposite when looking at the work from Gerner 2022, which is a Long Covid multi-omics profile. Specially confusing given that I am a Covid Long Hauler.

 

[Murph]

Would comparing be hard for some papers? The one from Fluge 2021 shows multiple subset of patient profiles. Given that ME/CFS is not an actual disease, but a syndrome, I think we are bound to see profiles that do not match, and just translate to similar symptoms. Summarizing them could be misleading.

I think it would be more useful if we looked at our own metabolite profile, and then see if they matched any of these papers. I haven't seen anyone here doing that yet. When looking at mine for example, I can see some things matching what Hanson 2018 found, like the disturbed Taurine pathway (I have borderline low levels, despite not being a vegan). On the other hand, I found exactly the opposite when looking at the work from Gerner 2022, which is a Long Covid multi-omics profile. Specially confusing given that I am a Covid Long Hauler.

1. I can't fully understand your comment, but are you worried that I'm trying to discredit these studies? That's not my goal. I'm trying to find the signal in the noise. The goal is to find areas of agreement, or, by bringing together more data, make subsets clearer. As for explaining differences between studies, It's certainly worth considering any extra data the papers have on who is in their me/cfs studies. And also how the samples were handled. I'm working on it.
2. The metabolites are found using a machine that's available to researchers only at this stage, so comparing our own results to the papers will not generate too much.

 

[Murph]

Here's another kind of output I'm hoping to make available: for every metabolite measured in Hanson's 2022 study, a bar chart showing each person's level of the metabolite.

This is an example of how such a chart would look for ADP (adenosine diphosphate, which is what's left after the energy system takes a phosphate off ATP; it is also a signalling molecule involved in platelet activation.)

 

[Murph]

Hey, I finally got one of these ideas sorted out and turned into a website. I'm proud of this, it's been a long road to be able to learn the coding to make this happen.

It has all the data from Germain and Hanson's amazing 2022 study on how MECFS patients metabolites change during exercise.

You can access it here, with a full 1100+ metabolites able to be accessed from the dropdown menu, dozens of patients and controls, and four measurements per person!

https://jasemurphy.shinyapps.io/Germainetal2022/

Any issues let me know (if you have suggestions for additional / improved functionality, please hold off! I'm exhausted! )