Research news from Bhupesh Prusty

As I recall, they did deep immune profiling of patient and HC blood. They then ran a Machine Learning classification algorithm on the samples. The algorithm separated into LC vs non-LC. This had 94% correlation with the patient's self-reported diagnosis (i.e included in study with or without active symptomatology).

Inclusion criteria for the Long COVID group were age ≥ 18 years; previous confirmed or probable COVID-19 infection (according to World Health Organization guidelines); and persistent symptoms > 6 weeks following initial COVID-19 infection. Inclusion criteria for enrollment of healthy controls were age ≥ 18 years, no prior COVID-19 infection, and completion of a brief, semi-structured verbal screening with research staff confirming no active symptomatology. Inclusion criteria for convalescent controls were age ≥ 18 years; previous confirmed or probable prior COVID-19 infection; and completion of a brief, semi-structured verbal screening with research staff confirming no active symptomatology.

UMAP embedding of study participants with all collected immunological features demonstrated a clear visual separation between people with Long COVID and those without.

UMAP is Uniform Manifold Approximation and Projection for Dimension Reduction. See UMAP Dimensionality Reduction — An Incredibly Robust Machine Learning Algorithm.
 
Yes, exactly. It's entirely circular. I think the following is right:

They had 99 people with self-reported symptoms consistent with Long Covid (and 40 uninfected healthy controls; 39 healthy convalescents and 37 healthy previously infected healthy care workers). The measure that the researchers found most predictive of Long Covid was slightly lowish cortisol. But we know that cortisol levels drop with lower activity, so that's not very informative.
Serum cortisol was the most significant individual predictor of Long COVID status in the model, and cortisol alone as a predictor achieved an AUC of 0.96 (95% CI: 0.92-0.99) in the data set (Fig. 6E). Notably, serum cortisol within the MY-LC study was highest in healthy (uninfected) controls, lower in convalescent controls, and lowest in participants with Long COVID (Fig. 6F). When tuning for accuracy, a threshold of 70.38 ng/mL obtained a maximum classification accuracy of 91.9%

They made a model that predicted who had Long covid, with cortisol levels being key part of that.

And then they say:
"Patient Reported Outcomes alone are sufficient to identify long COVID patients with 94% accuracy."

So, they are assuming that their model is perfect in identifying long COVID patients, and that, based on the categorisation that their model determined, 94% of the 99 people did in fact have Long Covid.

The thread for the Iwasaki study is here:
[Preprint] Distinguishing features of Long COVID identified through immune profiling, 2022, Klein, Iwasaki et al
 
Last edited:
Yes, I struggle with this and that's not even getting into the detail of ML algorithms. I guess there are two questions: is it valid to say that the two subgroups are so clearly different that this has to indicate an abnormal state (a "perfect model"); and whether this process is guided or unguided. Ie can you ask the machine the following?

Here is group of patient samples; each has "a bunch" of observational data.
Can you, the machine — with no knowledge of what any of these observations might mean — separate these samples into two or more sub-groups.
Let's say the machine reports back "there are two sub-groups in the samples you gave me", extremely widely divergent as defined by some valid distance-between-data mathematical construct.
The machine calls them 'Subgroup A' and 'Subgroup B'
You then do statistics referring back to your participants, which basically shows A = LC and B = HC per your/the patients' self reports (with maybe one or two incorrect).
You can then summarise the statistics as having 94% concordance between unguided machine classification and patient self-report.

That wouldn't be circular I think.

The alternative, that would be circular, would be guided classification: where you say here is 80% or 90% of our data-set; I'm telling you this subgroup is 'LC' and the other subgroup is 'HC'; can you make a classifier algorithm? We will then check this against the remaining 10 or 20% of our data-set to make sure it sticks.

I don't know anywhere near enough about ML to understand what they've done. There may be IT experts or scientists familiar with the techniques on the forum who could explain this better or further.
 
https://www.tlcsessions.net/episodes/episode-54-dr-bhupesh-prusty-molecular-virologist

He believes that he has found ‘the switch’.

Whilst searching for elevated markers in the bloods of patients with Long Covid and ME/CFS, Dr Prusty found the opposite – he believes that it is what is missing from the blood of these patients that may hold the key to treating the diseases.

And provide an actual biomarker!

In an introduction to his theories, this episode discusses viral reactivation, viral persistence and takes all of the symptoms and strands of Long Covid and puts forward a model that might explain all of it in one cohesive mechanism.

He very modestly thinks he may have found the mouth of the river where all the downstream manifestations that we suffer from have sprung from.

Dr Prusty hopes to publish his findings on this in the next month, and will present his model at the 12th Invest in ME Research Biomedical Research for ME Colloquium #BRMEC12 and 15th International ME Conference #IIMEC15 2nd June. Patients are welcome to attend.

The bad news is that although it may be possible to flip the switch back the longer we have been been on this journey the harder it is to reverse the down stream effects. Not impossible, just harder.
 
Here's an interview with Bupesh Prusty summarising his findings/thoughts so far.
Episode 54: Dr Bhupesh Prusty – Molecular Virologist
(Via TLC Sessions - Living with Long Covid.)
I hope you will find it illuminating. I think he's probably onto something fundamental to our disease. Let's hope so.

The interview with Prusty starts 8 minutes in. (It's general chit chat until then.)

https://www.tlcsessions.net/episodes/episode-54-dr-bhupesh-prusty-molecular-virologist
 
For anyone that can’t listen to the podcast, this bit is from 40 minutes in:

“We found that there is something which is normally there in every human being. In the patients, it starts to deplete inside the body. And it’s like a balance. You tilt toward one side, you start developing mild, tilt more, become moderate patient, you tilt completely, you completely lose it, become severe. Its the switch. It’s really amazing. The number of ME/CFS patients we have tested so far … The severe patients, they just completely lost this biomarker…”
 
For anyone that can’t listen to the podcast, this bit is from 40 minutes in:

“We found that there is something which is normally there in every human being. In the patients, it starts to deplete inside the body. And it’s like a balance. You tilt toward one side, you start developing mild, tilt more, become moderate patient, you tilt completely, you completely lose it, become severe. Its the switch. It’s really amazing. The number of ME/CFS patients we have tested so far … The severe patients, they just completely lost this biomarker…”

Thank you for sharing this! Too tired to go in and listen right now, but very interested in these updates on his research. It really would be amazing to have an identified biomarker. Though don't want to get my hopes up too much!
 
This is the first time I've felt genuine hope in two years. I hope prusty is really onto something here. I feel like if he was bullshitting at this point his reputation would be trashed?

Edit: By Bullshitting I mean exaggerating somehow, I'm not meaning to suggest prusty is lying.

And I am doing my best to temper my hopes with a healthy dose of realism.
 
Last edited:
For anyone that can’t listen to the podcast, this bit is from 40 minutes in:

“We found that there is something which is normally there in every human being. In the patients, it starts to deplete inside the body. And it’s like a balance. You tilt toward one side, you start developing mild, tilt more, become moderate patient, you tilt completely, you completely lose it, become severe. Its the switch. It’s really amazing. The number of ME/CFS patients we have tested so far … The severe patients, they just completely lost this biomarker…”
This could just be a downstream effect, the lower the activity (i.e healthy, mild, moderate, severe) the lower the levels of many "biomarkers". It may just be a pointer to how inactive patients are.
Until he tells us what it is we have no way of knowing. I would temper optimism until then.
 
This could just be a downstream effect, the lower the activity (i.e healthy, mild, moderate, severe) the lower the levels of many "biomarkers". It may just be a pointer to how inactive patients are.
Until he tells us what it is we have no way of knowing. I would temper optimism until then.
Indeed. I go back to my original thoughts which are that he should say nothing until he can say all of it. As John says it could simply be an illustration of lack of activity. Getting hopes up when you cant give all the info is cruel and irresponsible IMHO no matter how well meant.
 
Indeed. I go back to my original thoughts which are that he should say nothing until he can say all of it. As John says it could simply be an illustration of lack of activity. Getting hopes up when you cant give all the info is cruel and irresponsible IMHO no matter how well meant.

To be fair, he might have to publish some findings before getting ethical approval for further tests on inactive controls (tetraplegics? teenagers?), but the showmanship is definitely irksome.
 
Bhupesh says this new work is relevant to the 2020 paper that something in serum causes mitochondrial disfunction. This latest model shows what the thing in serum is.

He says: Take the thing and put it in cell and it causes disfunction in healthy cells, changing metabolism and source of energy and causing endothelial cell mito disfunction.

I thought it was a lack of the thing, not the thing itself that caused disfunction?

He also says that Robert Phair said his 2020 paper was “Nobel prize worthy”. The same paper that people with science backgrounds on this board thought was incoherent and really more of a hypothesis paper than a paper that proved anything.

I dunno.. I am skeptical about scientists who seem to court attention.
 
I have listened to the interview. It does sound hopeful that they are on to something real.
As a side comment when asked about microclots he thinks they are a downstream effect, so trying to clear them with apheresis will only be a temporary thing, and not get to the root of the problem.
 
Back
Top