Viren Jain

Wouldn't it be great if we could not only image large connectomic volumes but also completely reconstruct them? And if a whole mouse brain project didn't cost billions?

With the PATHFINDER preprint (www.biorxiv.org/content/10.1...), we preview a future where it doesn't have to.

We first imaged partially overlapping volumetric tiles arranged in two dimensions (2D) to obtain a first slab of LICONN data. We then used a conventional vibratome to slice off most of the imaged hydrogel layer, while placing the cut within the imaged slab. In a subsequent imaging round, we obtained an axially offset multi-tile volume situated more deeply in the tissue, featuring a continuous region of axial overlap with the first multi-tile volume. We then computationally fused the individual imaging volumes in 3D in a voxel-exact manner

Nice, thanks for pointing this out! "Saw the preview, don't have to see the movie" isn't always true. :)
The approach is simple and smart and allows one to sidestep the problem of how to losslessly subdivide a whole brain. Good.

An important addition to the Nature paper, beyond what was shown in the 2024 preprint, is 12 rounds of iterative imaging and sectioning of a LICONN volume, achieving 205 microns in axial extent (native scale) with manual tracing of axons: www.nature.com/articles/s41...

Light-microscopy-based connectomic reconstruction of mammalian brain tissue - Nature A technique called LICONN (light-microscopy-based connectomics) allows mapping of brain tissue at synapse level and simultaneous measurement of molecular information, thus enabling quantification of cellular properties and multimodal analysis of brain tissue.

Nature research paper: Light-microscopy-based connectomic reconstruction of mammalian brain tissue

https://go.nature.com/3YCLkuZ

New era for brain mapping, latest from Johann Danzl lab in @nature.com. Light-microscopy connectomics is at single-synapse resolution. Engineered hydrogel embedding + expansion, deep learning-based segmentation, and molecularly resolved reconstructions: axons, spines, connectivity, all visible.

We'll present our #ICLR2025 spotlight on ZAPBench this afternoon: 📍 Hall 3 #61!

GitHub - google-research/zapbench: The Zebrafish Activity Prediction Benchmark measures progress on the problem of predicting cellular-resolution neural activity throughout an entire vertebrate brain. The Zebrafish Activity Prediction Benchmark measures progress on the problem of predicting cellular-resolution neural activity throughout an entire vertebrate brain. - google-research/zapbench

Predict the zebrafish brain: github.com/google-resea...

ZAPBench ZAPBench evaluates how well different models can predict the activity of over 70,000 neurons in a novel larval zebrafish dataset.

⚡️ Excited to introduce ZAPBench, our #ICLR2025 spotlight: The Zebrafish Activity Prediction Benchmark measures progress in predicting neural activity within an entire vertebrate brain (70k+ neurons!)

Explore interactive visualizations, datasets, code + paper: google-research.github.io/zapbench

🧠🧪

Astounding achievement and even more astounding numbers here

TensorStore for High-Performance, Scalable Array Storage Posted by Jeremy Maitin-Shepard and Laramie Leavitt, Software Engineers, Connectomics at Google Many exciting contemporary applications of computer...

For more background on TensorStore check out the original blog post: research.google/blog/tensors... and reach out to me, jbms@google.com, or github.com/google/tenso... if you have any comments, questions, or suggestions.

New C++ API documentation has greatly improved the legibility of the C++ API: google.github.io/tensorstore/...

New batch read support has significantly improved efficiency when reading from sharded formats like zarr v3 and neuroglancer precomputed: github.com/search?q=rep... and many other general performance and efficiency improvements (e.g. github.com/google/tenso...)

OCDBT: multi-key atomic transaction support (non-distributed) · google/tensorstore@d77c943 PiperOrigin-RevId: 635977382 Change-Id: I993b6f012dba143485abff55e40a88fcd7759d8f

The new optionally cooperative distributed b-tree (OCDBT) driver has greatly increased performance on big workloads, and in 2024 we added transactional support to further optimize performance on object storage systems like GCS: github.com/google/tenso...

TensorStore (github.com/google/tensorstore) has had a huge 2024 and continues to power checkpointing for much of the largest ML work at Google, including Gemini, and support major Google Cloud customers training models on GCP. Read on for some specific 2024 updates!

Really moving, Rita. Thank you for sharing. I'm sorry for your loss.

I just put out v3.5.1 of NGLui, which is a python package for building and parsing data-driven Neuroglancer states. It's a big update, with cool new stuff like annotation tags (at least if you're using a bleeding edge NGL deployment) and much richer parsing.

This paper is incredible. EM level connectomics on a light microscope.

www.biorxiv.org/content/10.1...

How AI could lead to a better understanding of the brain Early machine-learning systems were inspired by neural networks — now AI might allow neuroscientists to get to grips with the brain’s unique complexities. Early machine-learning systems were inspi...

Can a computer be programmed to simulate a brain?
www.nature.com/articles/d41...

Mexico City is gorgeous!

why did you want to drill the glass?

omg

Would you like to see more housing in San Francisco? If not, cool, we'll find something else to agree on.

But if you do, then come party with me and others at the YIMBY homecoming on October 13! Grab your ticket here: bit.ly/45mGrHD

Been using @waymo.bsky.social throughout SF the past week. Cannot believe what they have achieved. What a revolution.