Lorenz Lamm

Attention maps and PCA visualisations comparing mode

In case you're missing my poster at #NeurIPS2025 about how I fine-tuned DINOv2 to ophthalmological images, here are some animations so you don't miss out!

🔗 Preprint: doi.org/10.48550/arX...
🔗 Code: github.com/peng-lab/rmlp
🧵1/9

Are you also often frustrated when DINOv2 puts very high attention on background patches, rather than cute fox heads?

@virtualhomo.bsky.social found an elegant way to regularize DINOv2 training using randomised linear algebra 🤯

Check out his thread or even his poster if you're at #NeurIPS2025.

Randomized-MLP Regularization Improves Domain Adaptation and Interpretability in DINOv2 Vision Transformers (ViTs), such as DINOv2, achieve strong performance across domains but often repurpose low-informative patch tokens in ways that reduce the interpretability of attention and feature...

I wrote a paper with @lorenzlamm.bsky.social, @marionjasnin.bsky.social, @tingyingpeng.bsky.social, F. Eckardt and B. Schworm and got accepted at #NeurIPS2025 and fun fact: 90% of viewers are enby vegans! 🤟

So if u fit there, u might wanna check it out! Maybe also if you don't. We're allies here <3

I'm pleased to announce 🍦 Icecream 🍨 v0.3!

New features include:
* Training on multiple tomograms (same training time, with linear increase in RAM) 🚀
* Logging and plotting of the loss function 📉
* The --scale option is now called --eq-weight for clarity 😉

We'd love to hear your feedback! 🙏🏽

You might have noticed lots of activity in the napari project recently! 🚀 We're grateful for a grant from CZI that's keeping us going, but grants don't last forever: we're trying to figure out sustainable long term funding. Read our blog post to find out how you can help:

napari.org/island-dispa...

✨Excited that the main project of my PhD is now available as a pre-print on #bioRxiv

Here, we used #CryoET to visualise mitochondrial proteostatic stress and together with SPA #CryoEM shed light into the functional cycle of the Hsp60:10 chaperone system. #TeamTomo

🔗 www.biorxiv.org/content/10.1...

Time for a thread!🧵 How different is the molecular organization of thylakoids in “higher” plants🌱? To find out, we teamed up with @profmattjohnson.bsky.social to dive into spinach chloroplasts with #CryoET ❄️🔬. Curious? ..Read on!

#TeamTomo #PlantScience 🧪 🧶🧬 🌾
elifesciences.org/articles/105...
1/🧵

🌱 Using ‘compelling’ methods, including #CryoET, researchers mapped spinach thylakoid membranes at single-molecule precision, revealing how photosynthetic complexes are organised and settling long-standing debates on chloroplast architecture.
buff.ly/j3TSIkn

A realistic phantom dataset for benchmarking cryo-ET data annotation - Nature Methods A standardized, realistic phantom dataset consisting of ground-truth annotations for six diverse molecular species is provided as a community resource for cryo-electron-tomography algorithm benchmarki...

Better ML for cryo-ET starts with better benchmarks.

We built a phantom cryo-ET dataset (~500 tomograms) + hosted a Kaggle challenge.

The result: community models beat expert tools.

Read more in the @natmethods.nature.com article that just came out:
🔗 doi.org/10.1038/s415...

Proud to share our latest paper. doi.org/10.1016/j.cr...

Through the dedication of @glynnca.bsky.social and @cryingem.bsky.social we report a thorough method to image molecular organisation within hippocampus tissue.

Structural biology in tissue is well and truly here!

@rosfrankinst.bsky.social

Ooh that's awesome! Great to hear the new version improved your segmentations :)

Final PhD paper now reviewed and published in JSB:X. I was happy with some great reviews that improved the paper! Thanks to Sander Roet for his help with the code base, and Remco Veltkamp and @fridof.bsky.social

We’re kicking off the DinoSphere Online Seminar Series! Join us for our first session with Karel Mockaer (Heidelberg) & Yong Heng Phua (OIST)

📅 1 July 9AM CEST
🔗 tinyurl.com/4mjaverj

Spread the word!
@protistwtmostest.bsky.social @ehehenberger.bsky.social @chandnibhickta.bsky.social&Norico Yamada

You want to start tomography? Solve structures inside cells? Reach Nyquist 😳 ? @phaips.vd.st and I have a website for you! tomoguide.github.io
You'll find a tutorial on how to reconstruct tomograms, pick particles and do subtomogram averaging, using different software!
Hope it will be useful !

Hey #TeamTomo,
Ever been in need of a tutorial about the fundamentals of cryo-electron tomography? From preprocessing raw frames to high-res subtomogram averaging?
That's why @florentwaltz.bsky.social and I made this website!

tomoguide.github.io

Follow the thread 1 /🧵
#CryoET #CryoEM 🔬🧪

🚀🔬🦠 Releasing 🤖Cellpose-SAM🤖, a cellular segmentation algorithm with superhuman generalization 🦸‍♀️. Try it now on 🤗 huggingface.co/spaces/mouse...

paper: www.biorxiv.org/content/10.1...
w/ @computingnature.bsky.social 1/n

MemBrain v2: an end-to-end tool for the analysis of membranes in cryo-electron tomography

Figure 1

Figure 2

Figure 3

MemBrain v2: an end-to-end tool for the analysis of membranes in cryo-electron tomography [updated]
Cryo-ET membrane analysis via deep learning.

We’ve updated our powerful MemBrain-seg tool for CryoET membrane segmentation! Plus, we’re introducing two new tools: MemBrain-pick for particle picking and MemBrain-stats for statistical analysis. Feedback is warmly welcome!

Check out the latest version of MemBrain, spearheaded by computation superstar @lorenzlamm.bsky.social ! It can segment, pick particles and give you metrics on everything!

📣Huge thanks to Simon, Hanyi, @lifeonthewedge.bsky.social, @florentwaltz.bsky.social, @wojwie.bsky.social, @kevinyamauchi.bsky.social, @alisterburt.bsky.social, Ye, Antonio, Sebastian, Fabian, @ja-schnabel.bsky.social and especially @cellarchlab.com & @tingyingpeng.bsky.social for incredible support

🤝 Feedback
If you feel like trying one of our modules or even the full pipeline, please let us know how it goes. We are happy for any feedback and would love to improve MemBrain v2 even further to make it as helpful for the community as possible.
🧵(6/6)

🔑 Usability
We focused on making MemBrain v2 smooth to work with: MemBrain-seg works with a single command line, while MemBrain-pick enables data-efficient training. We facilitate the transition between modules with several Napari functionalities like the 3D lasso to crop areas of interest.
🧵(5/6)

⚖️ MemBrain-stats
This module analyzes the spatial organization of particles on membranes. It takes the outputs of MemBrain-seg and MemBrain-pick to compute metrics like particle concentrations and geodesic nearest neighbor distances.
🧵(4/6)

⛏️MemBrain-pick
If you’re interested in localizing membrane-associated particles, please give MemBrain-pick a try. It enables efficient training of a model to localize particles on membranes and works with the Surforama plugin for interactive annotation in Napari.
🔗 github.com/cellcanvas/s...
🧵(3/6)

🎨 MemBrain-seg
This module allows out-of-the-box segmentation of membranes with just a single command line.
It’s based on a U-Net architecture, trained with a diverse dataset to enable generalization to many settings.

🔗 github.com/teamtomo/mem...
🧵(2/6)

🦠🧠 MemBrain update! 🧠🦠
We’ve updated our preprint! It now covers the full MemBrain v2 pipeline for end-to-end membrane analysis in #CryoET: segmentation, particle picking, and spatial statistics.
🔗 Preprint: doi.org/10.1101/2024...
🔗 Code: github.com/CellArchLab/...
🧵(1/6) #TeamTomo

Yo #TeamTomo, check out our updated #MemBrain v2 preprint. And better yet, give it a whirl on your #CryoET membranes! Please send us your feedback! 🧪🧶🧬🔬

Figure 4. MemBrain v2 end-to-end workflow detects periodic phycobilisome organization. A: Raw tomogram slice of EMD-31244. B: Out-of-the-box MemBrain-seg segmentation (light blue). C: A single membrane instance can be visualized in Surforama and manually annotated with GT phycobilisome positions (magenta). D: MemBrain-pick localizes particles (trained with data from C) on all membranes in the tomogram. E: MemBrain-stats computes Ripley’s O statistic using the positions from D with a bin size of 5nm. The distance between peaks (35 nm) was measured to estimate chain unit spacings.

We have updated our #MemBrain v2 preprint with a lot more details about the MemBrain-pick and MemBrain-stats modules, as well as some application examples!

Stay tuned for the upcoming thread by lead author @lorenzlamm.bsky.social! 🧠🧵

#CryoET #TeamTomo
www.biorxiv.org/content/10.1...

I've got bad news for #teamtomo: earlier today, our resident headcrab attacked @lorenzlamm.bsky.social, turning his Mem🧠 into 🧟‍♂️🧠

Nothing to worry about. I could backpropagate the attack and MemBrain's neurons are now more robust than before 💪