MECFS data analysis thread

When comparing blood metabolomics papers, note the preference for plasma over serum as described in Plasma Instead of Serum Avoids Critical Confounding of Clinical Metabolomics Studies by Platelets (2024, Journal of Proteome Research). It may be best to discard any serum studies, or at least sub-group them like male:female.

 

Great work @chillier !
i want to come back and respond to that but first I will post the data and code for the analysis I did above.

So here's a link for the Hanson 2020 data (clicking this will download a zip file to your computer) https://www.mdpi.com/2218-1989/10/1/34/s1
and here's a link to the code I used to compare it to Hanson 2022.

https://github.com/jasemurphy/mecfs/blob/main/Hanson_2020_vs_2022.R

 

1. This is so wonderful to see, thanks heaps for your hard work! my dream with this project was the many impressive people in this community would pitch in and it is starting already!

2. Is your topline takeaway from this that hanson 2020 and 2022 have no standout areas of agreement? Is that a fair summary? of course there are those metabolites that they agree on and which are significant but just as many that they disagree on which are significant. It would be excellent to see if any *other* studies have looked at those that they do agree on.

3. would you be willing to share your code? even just crtl+c, ctrl+v dumping it into a comment here would be useful to me!

4. did you use the D1-PRE data from Hanson 2022? and does it change if you look at the other timepoints?

 

This next chart compares Hanson 2020 to Naviaux 2017. There's not loads of metabolites in common but the ones I found in common do not show strong agreement. Note this chart has lipids and other molecules all mixed in.



here's the code for this one: https://github.com/jasemurphy/mecfs/blob/main/Hanson_2020_vs_Naviaux_2017.R

I imagine some people may view this analysis as basic but there's a method to the approach. Before we go to complex interrogation of the data it's worth asking it some very simple questions. So far it doesn't seem to have many answers. Showing that fact is a bit like reporting your null results.

 

Here's another one while I'm on a roll. Naviaux vs Fluge 2021. This is the Fluge paper where they run some unsupervised machine learning to create subsets. This chart has all the subsets bundled together but the next step might be to see how it looks if you plot each separately.





code for this one: https://github.com/jasemurphy/mecfs/blob/main/Naviaux2017_vs_Fluge2021.R

 

Jonathan Edwards's post here*, re TPPP1 gene, illustrates(?) a potential scenario i.e. a target gene/pathway is discovered and you run a metabolomics study on that group:

• untargeted i.e. measure all the metabolites you can (hypothesis free);
• targeted at e.g. pathway you consider should be relevant (hypothesis driven).

As per your comment, a hypothesis driven (i.e. targeted) metabolic study might e.g. be able to look at adjusting the procedure to increase sensitivity (detectability) of certain metabolites.

Still haven't seen a comment re limitation of Metabolon data (1000 metabolites) and how that could be addressed e.g. did Hanson do a study looking for a wider range of metabolites - is that the sort of thing we need?


*Jonathan - "Who knows? - we would need to find out. Maybe that the critical abnormal process in ME is so hard to observe because it involves supramolecular solid phase changes within brain cells at a level that at present we have no means to observe directly?"
https://www.s4me.info/threads/genet...022-hajdarevic-et-al.25070/page-3#post-411554

 

There are some great points here about the limits of knowledge. I want to describe how I see it.

We know most studies cover only a fraction of the metabolites in the body. Even when thousands are measured it's possible there's a smoking gun that we haven't measured or that science can't yet measure.

It's possible that the body fluids measured don't contain any evidence of illness at all. or it's possible the evidence is present only under certain conditions.

Nevertheless it's also possible there is a metabolite or two that will show up as elevated across several studies, hiddden only by statistical noise. That's what i'm trying to sort out by including everything, whether deemed statistically significant in an individual study or not. A perfectly useful answer will be, nope, you can't find anything that stands out at a metabolite level. The result of that might be to change the way scientists approach metabolomics in mecfs.

After metabolite-level analysis the second-best approach is pathway based analysis.

 

1) Thank you! I'm glad you think so

2) I think that's basically right, just because the ones they agree on are not significant after multiple test correction. I don't think it's necessarily problematic that they disagree on lots of things though - the ones they disagree on that are most significant are all drug metabolites and it's totally fine that those disagree.

3) Yep, no problem:

hanson2020.csv is the link you gave https://www.mdpi.com/2218-1989/10/1/34/s1 supplementary 1, saved as a csv. hanson2022_raw2.csv is https://insight.jci.org/articles/view/157621/sd/2 with the 'original' sheet saved as a csv. Then just set the working directory at the top and it should all run without issue hopefully. The plots it generates look slightly different to the ones I posted before, because previously I used the scaled data whereas now I am normalising myself with the raw data - I'm normalising slightly differently (by sample rather than by metabolite).

4) Yes I'm using D1-PRE, you can change to other time points from one line in the code (in the data preprocessing section).

 

I just want to point out that these kind of -omics studies are still coming! Bergquist and Armstrong are going to drop a really detailed study soon, with LOADS of datapoints taken at 20 minute intervals and a couple of thousand metabolites and lipids. I'm keen to wrestle the existing data into a format where we can see more easily whether the upcoming findings confirm prior signals, upset the apple cart or simply reaffirm that the whole field is too noisy.








Some of the findings in the video look quite good I will say.

 

It seemed from the Berquist presentation where those slides come from (Lisbon Conference Apr 4th 2024) that they really wanted to dive into the Lipids to get a better understanding about what exactly is happening. Low ceramides showed yet again!. I think he said that he was putting together an even more detailed metabolomic analysis that will have even more lipid identification.

What was interesting is that there are two arms to the study. Berquist used exercise, social activity, and a mental task to track how metabolites change over 8 hours and Armstrong will have a set of 3 controlled meals to see how diet affects the metabolites. These are in home studies to remove the stress of traveling and being at a research lab.

 

I just came across the data for Lipkin & Fiehn 2022 metabolite paper in case you haven't seen it @Murph
https://pdfs.semanticscholar.org/d2c1/82b950b2529e8ea1a678813bb142abdf2b86.pdf