iiinova — where your digital self keeps evolving

Brian Camley @diffusiveblob.bsky.social · 20/04/2026

PS Wei has put in a ton of work to make the code easy to install - it's just "pip install pyafv" and it's basically just as simple to run a AFV model as it is to simulate a bunch of random walks:

for _ in tqdm.tqdm(range(1000), desc="Active dynamics"):
diag = sim.build()
forces = diag["forces"]

active_velocity = v0 * np.column_stack((np.cos(theta), np.sin(theta)))

pts += (mu * forces + active_velocity) * dt

# Gaussian white noise
theta += np.sqrt(2 * Dr * dt) * np.random.randn(N)

sim.update_positions(pts)

210

Brian Camley @diffusiveblob.bsky.social · 20/04/2026

Want to simulate large nonconfluent tissues? Try the finite Voronoi model! 1) @wwang721.bsky.social and I show past implementations need a correction to avoid issues, and 2) we provide a new fast code that we'd like people to try! Preprint: arxiv.org/abs/2604.15481 code: github.com/wwang721/pyafv

1166

Brian Camley @diffusiveblob.bsky.social · 04/03/2026

An active material that's trying to contract can be guided by changing its friction with the environment. We predict some interesting examples, including patterns that can drive circular or linear motion of clumps of material. Work with Miller Fellow Cody Schimming: arxiv.org/abs/2603.03232

1151

Brian Camley @diffusiveblob.bsky.social · 10/02/2026

It's honestly been quite a while since I used the residue theorem! I think most of the applications in physics are in field theories? Here's something from Srednicki's QFT book:

Derivation of the Green's function for an oscillator from Srednicki's QFT textbook, https://web.physics.ucsb.edu/~mark/ms-qft-DRAFT.pdf - discussion from Eq. 7.12 to 7.14 in that PDF

100

Brian Camley @diffusiveblob.bsky.social · 13/11/2025

Is there an easy way to actually find an article for Nature Communications without a title if you know its article number? Nature's search has the same problem - if you look for article 1, you find all the other articles: www.nature.com/search/advan...

screenshot of nature's search feature that shows you can only enter "start page / article no."

020

Brian Camley @diffusiveblob.bsky.social · 30/10/2025

Lots of other predictions, tests. We are hoping this stimulates some further experiments to try to prove us wrong! In particular, our model would predict that cells should be attracted to a particular point in the device - and we don't see this immediately in the data (but can't rule it out).

Model showing SNR predicted to switch sign as a function of distance along a device so cells to the left are attracted and cells to the right are repelled, leading to a stable equilibrium

150

Brian Camley @diffusiveblob.bsky.social · 30/10/2025

Why does internalization matter for chemorepulsion? Internalization decreases the amount of bound receptor. If you inhibit it, you get more bound receptor - and you get attraction again!

Fig. 4B of the preprint shows SNR is mostly positive if CCR7 cannot internalize

131

Brian Camley @diffusiveblob.bsky.social · 30/10/2025

We then propagate error from the noise in the ligand-receptor binding to the noise in the response, and work out the signal-to-noise. The triangle is the 0-500 ng/mL experiment - repelled! The square is the 0-100 ng/mL experiment - attracted!

Fig. 4A of the preprint shows SNR switches from positive to negative to positive as a function of ligand concentration [L]mid in a linear gradient from 0 to 2 [L]mid

131

Brian Camley @diffusiveblob.bsky.social · 30/10/2025

Why does response switch from increasing to decreasing as you go toward larger chemoattractant (larger probability of bound receptor)? If A and I are nonlinear functions of the bound receptor, the ratio of the two R = A/I can easily switch between being increasing and decreasing- chemorepulsion!

Description of Fig. 3 AB from the text: "The form of Eq. (11) as a ratio of two nonlinear functions can lead to multiple switches between chemorepulsion and attraction. We show an example in Fig. 3A in which I undergoes a rapid transition between its basal level and a high level as pb increases (i.e. a large hI), and A slowly saturates as pb increases. We see that at small pb, I does not change much, while A increases – leading to an increase in R = A/I. Then, at the transition point where pb ≈ KI, I increases rapidly with pb, leading to a decrease in R. Then, as I saturates, while A continues to increase, R will increase again. We plot dR/dpb for this case in Fig. 3B, seeing that we have chemoattraction at small pb and large pb, but in the intermediate range
where dR/dpb is negative, there will be chemorepulsion; this
chemorepulsive region is shaded in blue throughout Fig. 3."

140

Brian Camley @diffusiveblob.bsky.social · 30/10/2025

Then we take an idea from earlier work on growth cones and assume bound receptor regulates the eventual response of the cell via a nonlinear feedforward loop.

Nonlinear feedforward loop with bound receptor activating A, I nonlinearly, and A activating R, I inhibiting R, and R the response - sets the direction the cell travels

130

Brian Camley @diffusiveblob.bsky.social · 30/10/2025

Because we know internalization is important, we start with the simplest possible model for ligand-dependent internalization, which does reasonably at capturing the timescale for the experiments.

Fig 1AB of the paper, 1A showing the model for ligand-dependent internalization, Fig 1B showing how the % surface receptor decreases with time in model and experiment

120

Brian Camley @diffusiveblob.bsky.social · 30/10/2025

Ten years ago, I saw a paper with some data that has bothered me ever since: B cells in a 0-100 ng/mL gradient of CCL19 are attracted to CCL19, but B cells in 0-500 ng/mL are repelled (see movie, ignoring the big clusters for now!). Why? Here's our model! doi.org/10.1101/2025...

1268

Brian Camley @diffusiveblob.bsky.social · 03/09/2025

5/ In principle, this makes the cell's "effective" dissociation constant adapt perfectly to changes in the ligand concentration - ensuring ~50% of receptors are always bound,

$When ligand concentration c is changed, allosteric protein responds, and this leads the fraction of receptors bound to adapt back to 0.5$

100

Brian Camley @diffusiveblob.bsky.social · 03/09/2025

4/ If an allosteric protein binds to the receptor, like in the first diagram, and this binding changes K_D to K_D/α, you want more of the allosteric protein to lower K_D to adapt to lower concentrations. So the idea is that bound receptors inactivate G and unbound receptors activate G.

$Diagram of feedback mechanism: G, which can bind to the receptor, switches to G*, which cannot. This switch is promoted by the bound receptor fraction, and the reverse by the unbound receptor fraction.$

100

Brian Camley @diffusiveblob.bsky.social · 03/09/2025

1/n New preprint: how eukaryotic cells could potentially adjust to new environments with perfect adaptation of their receptors (but why they probably might not). doi.org/10.48550/arX...

Diagram of "ternary complex" model with receptor switching between unbound, ligand-bound, ligand+allosteric protein bound, and only allosteric protein bound states

1135

Brian Camley @diffusiveblob.bsky.social · 29/08/2025

Not quite sure why @aip-publishing.bsky.social has scummy advertising that pretends to be a download link. This isn't ideal for a scientific society. (Brought to you by Chrome no longer supporting adblockers!)

J. Chem. Phys. article with ad "Download to use extension (Free)" that presumably links to malware

330

Brian Camley @diffusiveblob.bsky.social · 08/08/2025

I was curious about these failure modes. Surprisingly robust that you get wrong answers but easy to fix. I suspect LLMs are most useful when solutions are easy to check but hard to generate

ChatGPT transcript - also shows the code generated:

What's 8.8-8.11?

ChatGPT said:
8.8
−
8.11
=
−
0.31
8.8−8.11=−0.31

You said:
Can you write some code to check that and run it?

ChatGPT said:
The code shows that
8.8
−
8.11
=
0.69
8.8−8.11=0.69, not
−
0.31
−0.31.
I made a sign mistake earlier — 8.8 is larger than 8.11, so the difference is positive.

You said:
How many bs are in blueberry?

ChatGPT said:
There are 2 “b”s in “blueberry.”
(and it did this by generating code to count them)

492

Brian Camley @diffusiveblob.bsky.social · 28/07/2025

New work from Emiliano Perez Ipina + me: if cells respond to chemoattractant A and chemoattractant B, when do they go to source A vs source B? Or do they ever wander back and forth? We show a lot of different options are possible: arxiv.org/abs/2507.19341

Streamlines showing the direction cells exposed to two competing signals (A and B) travel

274

Brian Camley @diffusiveblob.bsky.social · 15/01/2025

Google scholar has a known bug you should look out for when generating bibliography entries. If you published a paper and earlier had a published abstract with the same title, Google Scholar will assume that the abstract is the correct version. This paper was not published in Biophys. J!

Clip of Google scholar entry for paper "Emergent collective chemotaxis without single-cell gradient sensing" falsely claiming it's published in Biophysical Journal

000

Brian Camley @diffusiveblob.bsky.social · 09/01/2025

I'll be giving the Biological Physics / Physical Biology seminar this Friday (Zoom) on how to connect protein motion on the membrane to how cells respond to an applied electric field! Includes some unpublished work- some of the most exciting in my career! sites.google.com/view/bppb-se...

1244