New COVID personal science thread: observations on viral load.🧵
Fun inside!
Viral load plummeted after 4 days of Paxlovid. I did some math* to estimate changes in viral load.
*somewhere between back-of-the-envelope and semi-quantitative modeling
x.com/famulare_mike/status/1894124001204867424
As you can see, I went from screaming hot viral load on symptomatic days 2 thru 5, to very low but still detectable on day 7. Along the way, on day 5, I lost much of my sense of taste and smell, but it started coming back yesterday and is not bad today.
x.com/famulare_mike/status/1894155423659561000
Because I'm annoyed we don't have app readers to turn rapid tests into semiquantitative readouts of viral load, even though it's obvious to everyone who uses them that you can read out relative viral load with the intensity (and time to positive), I did some analysis.
First, convert to grayscale and use a color picker to grab approximate absolute brightness data for the test line, the control line, and the test strip background. Put that in a spreadsheet and plot.
Second, normalize the test and control data by the background to get the excess darkness relative to the lighting and materials (and put it on a log scale). The control line is less variable now, although there is still a difference between flowflex (day 2,7) and iHealth.
So, we go one step further to normalize the test line darkness relative to the control. And now we have a nice time trend of antigen detected.
And since the y-axis units don't have an absolute scale anymore, let's also just plot it as adjusted darkness relative to peak. What we see by eye at the top is now clear: after a small decline in antigen detected over the first 5 days, day 7 was ~12x less intense.
From there, what does that mean for my viral load (and likely infectiousness). For that, we need some idea of how more quantitative measures like pcr ct and genome copies vary with lateral flow test antigen binding. I couldn't find one paper that does that for either test, sooo
So I stitched together some bits from a few papers to get the gist.
1) pmc.ncbi.nlm.nih.gov/articles/PMC11237673/pdf/spectrum.00073-24.pdf
2) pmc.ncbi.nlm.nih.gov/articles/PMC9031584/
3) nature.com/articles/s41598-021-02128-y
I'm too tired right now to document for twitter how the next few images came to be, but I'll post code eventually..
A careful look at this stuff finds that genome copy number is not linear in antigen test line intensity. Rather, depending on the dataset and the paper, you can work through either direct genomes vs intensity or genomes to ct to intensity to find...
genome copies is proproptional to color intensity to a power of between-ish 1.4 and 3. I think what's going on is intensity is linearly proportional to genomes at low signal, but nonlinear diffusion-binding-optical junk kicks in at higher concentrations, so studies differ.
And the dynamic range of rapid tests with pcr ct is well-known to be something like ct =15 (for the hottest infection ever) to ct ~30 for rapid test negative.
pmc.ncbi.nlm.nih.gov/articles/PMC9031584/
So anwyay, with that in mind, I can stich together estimates of the pcr-equvialent ct value I would've likely had, and the relative genome copies from peak. Here ya go!
This is cool, because I find it really useful to know that I'm something like 30-200x less infectious per minute of contact than I was a few days ago.
At onset, I could've been an epic superspreader if I hung out in a crowded space. I felt fine. I've mostly felt fine this whole time. If I wasn't very curious (and thought I might have flu), I'd never have known! Even with so much immunization, superpreading remains a thing.
Second, now I still have to be cautious around my family, but I am able to feel a lot more relaxed about my n-95 getting the job done. That takes a large mental load off.
Anyway, I wish we had standard tools for this. This kind of thinking would be way easier if it was just published for every test. And, I didn't do this analysis for mask viral load (aerosols), but nose/throat swab. It would be easy to make that correlation standard too.
We could know so much more about bespoke, personalized infectiousness! Which would make mitigation easier, more specific, and more palatable. I was hoping that future would come to pass 4 years ago. I hope for it still!
You can read the unrolled version of this thread here: typefully.com/famulare_mike/M7r3MoL
Mike Famulare
@famulare_mike
Like polio, Twitter lingers on. Learning to be a father, husband, and person. Systems epidemiologist, see Google Scholar. All tweets mine alone, not my employer
