Developing a blood cell-based diagnostic test for ME/CFS using peripheral blood mononuclear cells, 2023, Xu, Morten et al

Hutan said:
I agree, thank you. What percentage of the variation did your LDA model explain @chillier?
Click to expand...

I get values of 53% for LD1 and 47% for LD2 for proportion of separation respectively. My understanding is that it's not right to interpret these values as explaining the variation in the data overall, but as the proportion each LD is responsible for separating the different groups in this particular analysis. They add up to one here and also the figures in the paper so I think it's basically suggesting in my case LD1 is doing a bit more of the separation than LD2. It's just the relative contributions of the two LDs.

In a PCA we could see what is actually contributing to the most global variance in the data (if anything in particular). If what you're saying about different cell types having very different raman spectra, you might guess that most variance is actually explained by cell type.
 
I do think there could be some utility in this method in finding out which features are most responsible for separating the groups. Each of the features will have a loading value for each LD (or PC) which basically represent that feature's contribution to that LD. So imagine if you saw for example that the wavelength corresponding to tryptophan had a really high loading value compared to other wavelengths for LD1, that would be a big clue that trytophan is important.

You could go on and quantify it and test it (which to be fair they do anyway) and correct for multiple correction and so on.
 
chillier said:
I get values of 53% for LD1 and 47% for LD2 for proportion of separation respectively. My understanding is that it's not right to interpret these values as explaining the variation in the data overall, but as the proportion each LD is responsible for separating the different groups in this particular analysis.
Click to expand...
Ah, thank you.

For Figures 2 D and E that makes sense - the axes label percentages add up to 1. For Figure 2F, the LD1 percentage is 52% and the LD2 percentage is 31%, so 86%. And for Figure 2G it has LD2 at 30% and LD3 at 24%. How does that work?
Screen Shot 2023-06-23 at 10.34.47 pm.png
 
Hutan said:
Ah, thank you.

For Figures 2 D and E that makes sense - the axes label percentages add up to 1. For Figure 2F, the LD1 percentage is 52% and the LD2 percentage is 31%, so 86%. And for Figure 2G it has LD2 at 30% and LD3 at 24%. How does that work?
View attachment 19766
Click to expand...

I think that might be to do with the number of groups. It looks like the number of LDs you get is 1-number of groups. Since they're looking at 4 groups in figures F and G there's presumably 3 LDs and they're only plotting two of them. The total of all three would be 100% I believe
 
Andy said:
For those interested/concerned about how different preservation methods and time to process may affect samples then this paper might be of interest, Time dependent changes in the bioenergetics of PBMCs: processing time, collection tubes and cryopreservation effects 2023 Werner et al
Click to expand...

I see that we may be having similar friendly conversations away from s4me :D very exciting

To add my own experience: we did not see an effect of freezing duration on changes in viability (data in our biomarker paper), but didn't check for effects of freezing duration for the few seahorse experiments we did with PBMCs (I wrote those off after a few runs and didn't put much stock into it due to low signal to noise ratio, that data is in my initial experimental paper), so it is possible that it is affected as others suggest :). May be something here.

Emphasises the importance of using fresh cells where possible if they do not have the ability to turn over damage from freezing in culture (because PBMCs won't do this, being proliferatively asleep)
 
Last edited:
Oxford talks (open to members of University only)

Developing a blood cell-based diagnostic test for myalgic encephalomyelitis/chronic fatigue syndrome using peripheral blood mononuclear cells
Abstract:
Myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) is characterized by debilitating fatigue that profoundly impacts patients’ lives. Diagnosis of ME/CFS remains challenging, with most patients relying on self-report, questionnaires, and subjective measures to receive a diagnosis, and many never receiving a clear diagnosis at all. ME/CFS lacks a single sensitive and specific diagnostic test making the development of a simple test with the potential for early diagnosis a critical goal. Early diagnosis would enable patients to manage their conditions more effectively, potentially leading to new discoveries in disease pathways and treatment development.

Peripheral blood mononuclear cells (PBMCs) obtained from ME/CFS patients exhibited altered mitochondrial function, indicating a difference in energetic function when compared to non-fatigued controls. As ME/CFS may have a systemic energy issue, studying PBMCs may provide a good model for understanding the pathology affecting other organ systems. We hypothesized that single-cell analysis of PBMCs might reveal differences in ME/CFS compared to healthy and other disease groups. Raman spectroscopy is a non-invasive and label-free approach to probe molecular vibrations in a sample, and when combined with confocal microscopy, it can interrogate individual cells. A single-cell Raman spectrum (SCRS) is a phenotypic fingerprint of all biomolecules in that cell and could potentially differentiate between various cell types and give insights into underlying biology

In this study, we utilized a single-cell Raman platform and artificial intelligence to analyze blood cells from 98 human subjects, including 61 ME/CFS patients of varying disease severity and 37 healthy and disease controls. Our results demonstrate that Raman profiles of blood cells can distinguish between healthy individuals, disease controls, and ME/CFS patients with high accuracy (91%), and can further differentiate between mild, moderate, and severe ME/CFS patients (84%). Additionally, we identified specific Raman peaks that correlate with ME/CFS phenotypes and have the potential to provide insights into biological changes and support the development of new therapeutics. This study presents a promising approach for aiding in the diagnosis and management of ME/CFS, and could be extended to other unexplained chronic diseases such as long COVID and post-treatment Lyme disease syndrome, which share many of the same symptoms as ME/CFS.
Click to expand...
 
Merged thread

https://onlinelibrary.wiley.com/doi/10.1002/advs.202302146

Research Article
Open Access
Developing a Blood Cell-Based Diagnostic Test for Myalgic Encephalomyelitis/Chronic Fatigue Syndrome Using Peripheral Blood Mononuclear Cells

Jiabao Xu, Tiffany Lodge, Caroline Kingdon, James W. L. Strong, John Maclennan, Eliana Lacerda, Slawomir Kujawski, Pawel Zalewski, Wei E. Huang, Karl J. Morten
First published: 31 August 2023

https://doi.org/10.1002/advs.202302146

Abstract

Myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) is characterized by debilitating fatigue that profoundly impacts patients' lives. Diagnosis of ME/CFS remains challenging, with most patients relying on self-report, questionnaires, and subjective measures to receive a diagnosis, and many never receiving a clear diagnosis at all.

In this study, a single-cell Raman platform and artificial intelligence are utilized to analyze blood cells from 98 human subjects, including 61 ME/CFS patients of varying disease severity and 37 healthy and disease controls.

These results demonstrate that Raman profiles of blood cells can distinguish between healthy individuals, disease controls, and ME/CFS patients with high accuracy (91%), and can further differentiate between mild, moderate, and severe ME/CFS patients (84%).

Additionally, specific Raman peaks that correlate with ME/CFS phenotypes and have the potential to provide insights into biological changes and support the development of new therapeutics are identified.

This study presents a promising approach for aiding in the diagnosis and management of ME/CFS and can be extended to other unexplained chronic diseases such as long COVID and post-treatment Lyme disease syndrome, which share many of the same symptoms as ME/CFS.


Click to expand...
 
Last edited by a moderator:
chillier said:
I've modelled this in R here using data with no signal and only random noise.

In this paper they have about 1000 features (readings for 1000 different wavelengths) over 1000s of cells - so its dimensionality is high. I've generated a dataset for 1000 'samples' each with 1000 'features.' The dataset is populated with random decimal values between 0 and 1, so there is no pattern only noise. I've then assigned each of the samples randomly to a group number 0, 1 or 2 to emulate groups of (controls, ME or MS).

Here is a scatterplot of features 1 and 2, each dot corresponds to a sample. You can see there's no pattern:
View attachment 19763

I've then split up the data into two parts, 70% of the samples will be used to train an LDA model to predict the groups from the data, and the remaining 30% will be used to test it. When you plot the first two LDs from the training data you can see it separates the groups amazingly - based off of absolutely no real signal at all. I was surprised at just how strongly this resembles the plot in the paper:
View attachment 19764

Then when you go on to use the trained model to predict the groupings on the test data you can see it can't do it at all:
View attachment 19765

here's the R code if anyone wants to retry:
Click to expand...
This is brilliant work. I’ve always wondered how a random data comparison would look for many different ME studies.

clearly, not very good in this case.
 
Last edited:
Seems like the big problem here is that they divided their data into 80% training / 20% test, but they divided data from cells (2155 cells from 98 patients and controls) rather than the people. So it doesn't seem surprising that their ML algorithm, which was essentially trained to distinguish cells from three different groups of people, is good at distinguishing cells from those three groups of people. What, if anything, that has to do with ME seems like an open question.

I guess they'd need samples from a lot more patients and controls to do the same separating the data by person instead of cell, but that seems like the obvious need here.
 
LarsSG said:
but they divided data from cells (2155 cells from 98 patients and controls) rather than the people. So it doesn't seem surprising that their ML algorithm, which was essentially trained to distinguish cells from three different groups of people
Click to expand...

Yes, exactly, good point. It's not just the machine learning but it looks like they've also done that where they compare and t test differences between metabolite levels. Comparing cells not individuals - it would just take one individual with lots of differing cells to make the entire cohort appear different.
 
ScienceAlert New Blood Test For Chronic Fatigue Syndrome Has 91% Accuracy

quote:
The blood test differentiates between the properties of a type of blood cell called peripheral blood mononuclear cells (PBMCs) in people with and without ME/CFS, using a technique called Raman spectroscopy and an artificial intelligence (AI) tool.

Previous studies have suggested PMBCs from people with ME/CFS have reduced energetic function; results which fit with an emerging theory that the condition is one of impaired energy production.

Building on their pilot study, and the research suggesting PBMCs are perturbed in ME/CFS, Xu and colleagues tested their diagnostic approach in nearly 100 people: including 61 individuals with ME/CFS, 16 healthy controls, and 21 people with multiple sclerosis, an autoimmune disorder that has many similar symptoms to ME/CFS.

If the blood test could distinguish between people with ME/CFS and those with MS, as well as healthy folks, then it might bode well for its use in differentiating ME/CFS from other illnesses, such as fibromyalgia, chronic Lyme disease, and long COVID.
 
JemPD said:
Would love to get your take on this @Jonathan Edwards if you have time/energy
Click to expand...

For me the basic problem lies in gearing up a study with the intention of producing a test that gives high specificity and sensitivity, on the basis that this would be useful in diagnosis.

Until we have some data showing that some single relevant measure is consistency different from normal in at least a substantial number of PWME I don't see progress. In other words I would forget about machine learning.

The other problem is that peripheral blood mononuclear cells (PBMC) are not a good cell type to test if we are interested in energy metabolism. PBMC are a heterogeneous mixture of cells mostly in a very inactive state. PBMC might reflect some general defect, so should not be dismissed, but they are also very likely not to, or to show something spurious secondary to altered daily activity.
 
Jonathan Edwards said:
For me the basic problem lies in gearing up a study with the intention of producing a test that gives high specificity and sensitivity, on the basis that this would be useful in diagnosis.

Until we have some data showing that some single relevant measure is consistency different from normal in at least a substantial number of PWME I don't see progress. In other words I would forget about machine learning.

The other problem is that peripheral blood mononuclear cells (PBMC) are not a good cell type to test if we are interested in energy metabolism. PBMC are a heterogeneous mixture of cells mostly in a very inactive state. PBMC might reflect some general defect, so should not be dismissed, but they are also very likely not to, or to show something spurious secondary to altered daily activity.
Click to expand...
thanks for that
 
Jonathan Edwards said:
For me the basic problem lies in gearing up a study with the intention of producing a test that gives high specificity and sensitivity, on the basis that this would be useful in diagnosis.

Until we have some data showing that some single relevant measure is consistency different from normal in at least a substantial number of PWME I don't see progress. In other words I would forget about machine learning.

The other problem is that peripheral blood mononuclear cells (PBMC) are not a good cell type to test if we are interested in energy metabolism. PBMC are a heterogeneous mixture of cells mostly in a very inactive state. PBMC might reflect some general defect, so should not be dismissed, but they are also very likely not to, or to show something spurious secondary to altered daily activity.
Click to expand...
Why didn't Morten look for a single measure to differentiate PWME from healthy people? Is this a step towards that? Could Decode ME participants be recruited?
 
You must log in or register to reply here.
Back
Top Bottom