Combinatorial protein barcodes enable self-correcting neuron tracing with nanoscale molecular context https://www.biorxiv.org/content/10.1101/2025.09.26.678648v1

Update From Doctors Without Borders: "This is one of the worst situations I have seen” ... "It really is quite unique in terms of the volume of affected people, the level of suffering and also the underfunding of the international response” www.ft.com/content/d331...

Wow, VERY interested to see this paper!

Settle a debate for me: If you could hook up a fruit fly brain to a completely new set of end effectors, could it learn to use them for goal-directed behavior? To what extent is an insect brain evolutionarily specialized for controlling its body? #FlyBCI

How to make frontier AI models safer and accelerate drug discovery with the connectome. We are featured in this
@ifp.bsky.social article by @adammarblestone.bsky.social & @andrewcpayne.bsky.social

ifp.org/mapping-the-...

The latest on this humanitarian catastrophe, one of several happening in the world right now, yet strangely neglected in the public discourse: www.theglobeandmail.com/world/articl...

There is a famine currently affecting 640k children in Sudan, driven by military action funded by a key US ally in the Middle East. It's been widely reported in the media since 2023. Yet no condemnation by public figures, no demonstrations in the US. Why does no one care?

Amazing work

(1/7) New preprint from Rajan lab! 🧠🤖
@ryanpaulbadman1.bsky.social & Riley Simmons-Edler show–through cog sci, neuro & ethology–how an AI agent with fewer ‘neurons’ than an insect can forage, find safety & dodge predators in a virtual world. Here's what we built

Preprint: arxiv.org/pdf/2506.06981

Shortly after MICrONS demonstrated initial scale-up of volume EM, I think in 2017, I desperately tried to convince NSF bio division leadership (then Olds and Deshler) to launch a mid-scale infrastructure project focused on comparative connectomics across species. They were interested, but no budget.

The writing's on the wall that the future of neuroscience is team science. The key question is how to balance support for "top-down" vs "bottom-up" collaborative research efforts. I agree teams of early career researchers deserve greater agency. Thoughtful piece by @neuralreckoning.bsky.social 👇

This 👇 is a major advantage of optical circuit mapping approaches over non-destructive (i.e. non FIB) serial section EM. Fewer failure modes that prevent tracing across distinct imaging volumes.

For example, in my experience with Claude 3.7 Sonnet, type errors are particularly common when passing matrices around.

GD requires so much data to train foundation models, however. Data we don't have in neuroscience. People are trying, but what if the time required to collect all the physiology data needed to "properly" train a whole brain model is O(decades)? (or longer)

50/50 would be the best odds I've had on anything I've tried for the last 10 years 😉 I'd be stoked to test this hypothesis in a "small" system with well-characterized dynamics, like STG. If that works, move to more complex systems.

It's not obvious to me that getting "all the parameters correct" is the right goal to set for a mechanistic model. At a molecular level, brains are highly dynamic and parameters are constantly in flux. Maybe there is a very large space of parameterizations that recapitulate the same dynamics?

If dynamics are fully determined by system parameters (structure + cellular+molecular properties), shouldn't the model exhibit "normal" function if initialized to any reasonable activity state? Why do we need to observe any physiology in advance?

If dynamics are fully determined by system parameters (structure + cellular+molecular properties), shouldn't the model exhibit "normal" function if initialized to any reasonable activity state? Why do we need to observe any physiology in advance?

The answer, I am convinced, is yes, and all the technologies needed to deeply phenotype a whole brain exist today. This is a viable path to whole brain emulation, with no physiology or behavioral experiments required.

Thought problem: If you could deeply phenotype a whole brain - mapping all ultrastructure and connectivity, plus in situ proteomics, metabolomics and transcriptomics - would that be sufficient to constrain a model of its function under all conditions? What's the minimal data needed for emulation?

Gratified by today's news of a $155M seed round for DNA data storage newco Atlas, built on technologies developed under the IARPA MIST program. It validates my "if we de-risk it investors will come" thesis. Also means my public service has now driven >$1B of economic activity 🙂

I've said it publicly before and I'll say it again: science awards that preferentially recognize individuals are an anachronism. This just doesn't reflect how much of today's impactful science gets done. 4/