Amatica Patient-centred chronic disease research

What are their criteria for determining an ↑ or ↓ result?
From what I understand, Amatica compares gene expression levels with those of a healthy control group, after normalising the data. Their analysis published in July included 159 patients and 85 healthy controls, matched for age and sex.

The differences seem to be assessed using z-scores, among other things. They also mention using statistical significance thresholds with corrections for multiple comparisons in their group analyses.

However, I couldn't find the exact threshold they use for the ↑ and ↓ arrows in our individual reports. I don't know whether it's simply based on a particular z-score or whether they apply other criteria. That's a good question to ask Jack, because it makes quite a difference when interpreting the results.

By the way, if you'd like, I can send you my login details and my patient report by private message so you can have a look yourself. I trust the people on the S4ME forum 100%.
 
However, I couldn't find the exact threshold they use for the ↑ and ↓ arrows in our individual reports. I don't know whether it's simply based on a particular z-score or whether they apply other criteria.
I could download a csv file from my account with the exact thresholds (upper and lower bounds) and my own z-scores for 3000 genes.
 
Interesting. Mine looks completely different with lots of stuff either “elevated” or normal, perhaps, because I am a female.

View attachment 34962
I don’t feel like this chart with grouped z-scores reflects anything meaningful honestly. Maybe because I don’t understand how they calculated my 1 grouped mean z-score for, for example, 115 genes in one group and then somehow calculated the upper and lower bounds for the whole group?… this doesn’t make much sense… for example, in my case 13 genes are upregulated in the autoimmune group, but somehow after grouping the z-scores the average result is just closer to upper bound on this chart.

They mention this on the website:
Category-level patterns reflect the overall direction across a group of genes, not individual gene values. All individual genes can be within range while the category pattern looks elevated, or vice versa

I still don’t get it.
 
Last edited:
Yes, 300 steps per day
That's exactly the number of steps I take every day. I've been bedridden since March 2025, but I think the illness had already started developing in January 2022, or perhaps even a little earlier.

By March or April 2023, I was already struggling badly with PEM, and exercising was becoming almost impossible. My body simply couldn't handle physical exertion the way it used to. I would get out of breath very quickly, with a feeling of impending doom and a strange pressure in my abdomen. Basically, symptoms I'd never experienced before.

Anyway, I find it quite surprising that our results are so different. Yours are almost normal, while mine are absolutely terrible, even though we seem to be at a fairly similar level of severity today.
 
From what I understand, Amatica compares gene expression levels with those of a healthy control group, after normalising the data. Their analysis published in July included 159 patients and 85 healthy controls, matched for age and sex.
The differences seem to be assessed using z-scores, among other things. They also mention using statistical significance thresholds with corrections for multiple comparisons in their group analyses.
However, I couldn't find the exact threshold they use for the ↑ and ↓ arrows in our individual reports. I don't know whether it's simply based on a particular z-score or whether they apply other criteria. That's a good question to ask Jack, because it makes quite a difference when interpreting the results.

I found these on Twitter that suggest the ranges for individual markers are the 5th-95th percentiles from the control group. And then my best guess is for the group stuff in the circle heatmap thingy, they calculate the mean of z-scores for that group and compare it to the IQR (i.e. 25th-75th percentile) of the control group. Idk how/when they correct for multiple comparisons tho - other than that it would presumably be using False Discovery Rate method.



 
I was just looking over their page again. There’s no standards here. It’s hard to tell if there is even controls per batch. There could be widely varying batch effects.

There’s no governing body or publication of how any of this works. You have to blindly test they are doing everything the same and “research quality”. I worked in a testing. Trust me there can be some wild effects in processing depending on who, what, and when things were processed. You even see this in small sample papers where you think it’d be easy with small manageable amounts of samples. There’s actually whole companies now based around A.I. glasses in the lab that record the scientist movements and classify differences for batch effects.

That is though to say this should be a pretty standard protocol, so I hope they have controls and rigid SOPs.

Also going in batches of 20 implies they might freeze samples at one point but you will not know. I get what they are trying to do but this just seems insanely expensive and noisey
 
Last edited:
3252.webp
amatica's new phase


Amatica started as a patient community project, built around a belief that patients with complex diseases such as Long COVID and ME/CFS need precision medicine that matches treatments to biological signatures to improve unpredictable treatment outcomes.

The problem​

Currently, millions of patients are left without viable care and are forced to experiment with their own health, taking off-label medications in hope of a positive response. While a minority of patients find treatments they benefit from, far more have no response and some are left permanently worse. Our team has experienced this ‘treatment roulette’ first hand.

As patients ourselves, we didn’t have years to wait for small clinical trials that average patients together to advance.

Our solution​

We believe that the future of care will be shaped by precision medicine and individual agency.

We offer patients the next generation of biological tests, combined with deep characterisation of disease profiles and states.

We are building the richest multimodal dataset, designed for analysis using state-of-the-art machine learning and AI, with the aim of improving the lives of millions of patients faster than was previously possible.

Today we're taking our biggest step towards our goal. We've closed a multi-million dollar seed funding round with the goal of building the leading precision medicine company in chronic neuroimmune disease, starting with Long COVID and ME/CFS.

Building the team​

After volunteering since our launch in 2025, our team is now full-time, and we're expanding. We're hiring leading computational biologists, data scientists and machine learning researchers to work on one of the richest patient datasets in this field. If that sounds like you, see our open roles.

Making the test affordable​

Without the ability to subsidise cost, our test was out of reach for many. We're now putting a significant share of our funding into subsidising it.

From today, our RNA test costs $495. We believe this can accelerate building the data scale needed to make an impact in patient care. This subsidised price is available to the first 500 new patients who consent to research.

Our best analysis yet​

One of Amatica’s goals is to make the richest and most complex biological data understandable to everyone.

Every patient now receives an intuitive summary, built from personalised pathway analysis and clustering across the 60,000+ gene expression measurements from your RNA sequencing.

Your report provides insight into the states of key biological systems (including mitochondrial function, immune states, vascular function and inflammatory signalling) and where your profile sits compared with other patients and reference controls.

As our researchers identify new subgroups and associations, those analyses are added to your portal at no additional cost, so you can see where you fit as the science develops.

Building a multi-omic dataset​

We're building a multi-omic dataset that brings different types of biological information together to power state-of-the-art machine learning research.

With consent, every new patient's blood draw will also provide five additional sample types, banked for future proteomic, metabolomic, genomic, epigenomic and extracellular vesicle analysis. Our test today is RNA sequencing only. The banked samples mean future analyses can be run without another blood draw and data can be combined.

Proteomics will be added next as a paid add-on, with a 1,000-protein panel. Over time, we'll integrate these layers and apply machine learning and AI to find signatures linked to disease states and biological subgroups.

What we're seeing so far​

Early analysis of our cohort points to shifts in B cell states, mitochondrial pathways, stress hormone responses and antiviral signalling. This builds on our preliminary classifier research covered in our recent blog post.

Our existing patients​

The people who supported Amatica at full price made all of this possible. We've set aside funds so every existing RNA patient who has consented to research receives a second longitudinal RNA test free. A second sample lets us see how biology changes over time and with disease states, and longitudinal data like this is some of the most valuable for research. Every existing patient also receives the new pathway and clustering analysis.

Blood collection and shipping​

We’ve simplified blood collection. The blood draw is now included in the price rather than an extra cost, and no centrifuging or additional processing is required. We can connect you with our partner mobile phlebotomists and support every step of collection and shipping. In the US our partner phlebotomist will handle return shipping for you. RNA sequencing is now carried out by our new CLIA/CAP accredited lab partner in the US.

Our vision​

As severely affected patients ourselves, we see advancing care as our only path to regaining our lives.

Next-generation tests, the maturation of precision medicine and the arrival of AI are converging. We believe this is the moment when decades of impasse in chronic disease care can be overcome.

Our model is designed to leverage this convergence to improve the lives of millions of patients within years, not decades.

Thank you to everyone who has been part of our mission so far and we are excited for this next phase
 
Next lab is CLIA compliant at least. It’s be interesting to run the first cohorts through this lab and see if it’s similar.

I still think a model where anyone can send in samples and it’s up to the patients to identify what their symptoms/disease is not viable.

Look at the GWAS work when they have integrated larger cohorts into the DecodeME sets, million veterans & general ICD identification. Very few hits replicated in the GWAS and that was with doctors doing diagnosis criteria not patients. You’re going to be comparing noise and I don’t even think next years frontier AI model can fix that. Garbage in garbage out
 

Attachments

  • IMG_4023.webp
    IMG_4023.webp
    53.5 KB · Views: 7
The tubes contain an RNA stabilizer.

According to the manufacturer's specifications, RNA remains stable for up to 3 days at room temperature after blood collection.

The samples are shipped via DHL Medical Express and then stored at approximately −80°C upon arrival at the laboratory.

I can personally confirm that this is how the process was carried out, as I went through it myself.
 
Batch effects for RNA is a real thing:



They need to come out and say how they control for this BEFORE sick patients pay the money. I also disagree patients should pay for this themselves but that’s another argument.
 
I'm not a scientist, so I can't really comment on batch effects. However, can't RNA expression vary considerably from one month to another, depending on a wide range of external factors? If that's the case, can these analyses really be considered a reliable tool for guiding individual treatment decisions?
 
I’m sorry but even the changing of seasons you will have different RNA expression between the same people? I could be completely wrong here:


There’s just SO much to control if you’re comparing everything
 
GWAS, like decodeME, controlled for heritage as well as far as I can tell this does not. Sorry to go off but this just seems so unconvincing as a real path and worse charges patients large sums of money.


I think a good litmus test without even being a scientist is why hasn’t this been done before for every disease? Seems pretty easy. From a capitalist perspective you could easily identify any disease and any pathway and find new targets netting you a competitive advantage. A couple million for some RNA reads is a drop in the bucket for pharma or the NIH. Maybe because it’s too good to be true
 
Last edited:
Oh no, don't apologize! Quite the opposite, I welcome critical thinking. I want to understand, and I find all of this genuinely interesting.

So does that mean RNA analysis isn't really a reliable way to better understand one's own disease?
 
My personal opinion no, especially currently. The main idea here is that with enough samples, power, the noise goes down.

However you can increase power but without controls (per batch, heritage, season, age, sex, disease. even previous infections etc everything that makes humans different when you do a massive fishing trip of RNA expression) you do not decrease bias. You’re just testing so much that’s is so variable.

I’d love jack to explain how they plan to tackle this other than saying big data or AI will fix it in the future.
 
So does that mean RNA analysis isn't really a reliable way to better understand one's own disease?

I would be 95% sure that is the case. I am baffled by the extent to which biomedical research has been entirely taken over by this sort of mindless measuring that has never provided anything useful in any well characterised disease as far as I know. You might as well go bus spotting as a way of discovering whether the stock market is going to crash. Most of the measurements are of things that vary for reasons nothing to do with what you are interested in.
 
Back
Top Bottom