Excited to share our new paper (CVPR 2026 π): "MuViT: Multi-Resolution Vision Transformers for Learning Across Scales in Microscopy" which enables local predictions to use global context.
Great work led by @albertdm.bsky.social & another fun collab w @gioelelamanno.bsky.social! @scadsai.bsky.social
Join us in building foundational datasets and models for microscopy with @jakobtroidl.bsky.social and many others
AI Engineer Pre-Training: lnkd.in/eyEEV5nm
AI Engineer Post-Training: lnkd.in/eShVQ3y9
Data Engineer Data: lnkd.in/eCvm58xH
Data Engineer Pipelines: lnkd.in/ec6qGJkX
ai.hhmi.org
Today I'm announcing a new digital textbook ππ₯οΈ, "An Introduction to OME-Zarr for Big Bioimaging Data".
ome-zarr-book.readthedocs.io
A nicely written opinion piece comparing small molecule vs genetic screening
www.sciencedirect.com/science/arti...
Glad to see this work finally published! We used nELISA in combination with CellPainting for our exploration of inflammasome inhibitors.
Job alert!
We are looking for a bioimage analysis to work in an image analysis core facility in beautiful Paris. Can I ask you to share this opportunity with your networks?
See also on the forum:
forum.image.sc/t/research-e...
mitochondria from bipolar patients are closer to the nucleus in these images; control patients' are spread out further
15 years in the making, we confirmed that mitochondria - the powerhouse of the cell - have an unusual localization in patients who experience psychosis (including schizophrenia and bipolar disorders). Youβll never guess what kind of patient cells we used to make this discoveryβ¦ π§΅
Very cool! Why call it CellTransformer, though, if the underlying inductive bias is organization in tissue? So confusingβ¦
Happy to share that ShapeEmbed has been accepted at @neuripsconf.bsky.social π SE is self-supervised framework to encode 2D contours from microscopy & natural images into a latent representation invariant to translation, scaling, rotation, reflection & point indexing
π arxiv.org/pdf/2507.01009 (1/N)
Morph Map is now published in Nature Methods. Excited to see what the community discovers with this resource mapping ~15,000 human genes!
rdcu.be/ezGre
One thing that really bothers me with the new "virtual cell" terminology is that it is currently largely focused on a very narrow definition of models that can predict effects of trans perturbations (gene dosage, drugs etc) on gene expression. 1/
Is your lab/institution willing to allocate budget and time towards outreach training and events? Iβm sure grad students would love the opportunity to earn some extra cash to show off their research.
CERN has been doing this very effectively for a long time cerncourier.com/a/your-guide...
Very cool results! Just started exploring the data and seeing some oddity: github.com/broadinstitu...
Is any of the authors still monitoring the repo?
![(a) Human U2OS cells treated with dimethyl sulfoxide (DMSO) and stained using the Cell Painting assay, which employs six dyes in five channels to label eight cellular compartments. The top row (from left to right) shows mitochondrial staining; actin, Golgi, and plasma membrane staining; and nucleolar and cytoplasmic RNA staining. The bottom row (from left to right) displays endoplasmic reticulum staining, DNA staining, and a montage of all five channels (from Cimini et al. [21]). (b) Thousands of features are extracted from each segmented cell in microscopy images of wells. A learned function f(x) (CytoSummaryNet) aggregates this data into a single feature vector: the sampleβs profile. (c) An in-depth look at the model architecture used in this study. The model consists of three elements: a function Ο(x), which maps the input data from βD to βL space, a summation, which collapses the cell dimension, and Ο(z), which maps the collapsed representation from βN to βL space. (d) During training, replicate compound profiles are forced to attract each other (green arrows) and simultaneously repel every other compound (red arrows) in the learned feature space. Here, all forces are drawn for a single profile of compound B.](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:raswr6mjt7sbppdo6kcce4mz/bafkreicppuq2utmxkx5kpaggndigq7qcat6ppdqqajxhphpyi53ydlbbeq)
(a) Human U2OS cells treated with dimethyl sulfoxide (DMSO) and stained using the Cell Painting assay, which employs six dyes in five channels to label eight cellular compartments. The top row (from left to right) shows mitochondrial staining; actin, Golgi, and plasma membrane staining; and nucleolar and cytoplasmic RNA staining. The bottom row (from left to right) displays endoplasmic reticulum staining, DNA staining, and a montage of all five channels (from Cimini et al. [21]). (b) Thousands of features are extracted from each segmented cell in microscopy images of wells. A learned function f(x) (CytoSummaryNet) aggregates this data into a single feature vector: the sampleβs profile. (c) An in-depth look at the model architecture used in this study. The model consists of three elements: a function Ο(x), which maps the input data from βD to βL space, a summation, which collapses the cell dimension, and Ο(z), which maps the collapsed representation from βN to βL space. (d) During training, replicate compound profiles are forced to attract each other (green arrows) and simultaneously repel every other compound (red arrows) in the learned feature space. Here, all forces are drawn for a single profile of compound B.
Taking pictures of cells with a microscope, then extracting thousands of features from them is uncannily effective for quantifying cell state, esp. for genes and chemicals (e.g., Cell Painting). But we often average the rich single-cell data to simplify analysis. Can we do better?
#bioML π§ͺ
1/n
Excited to present our spotlight paper at #NeurIPS!
MOTIVE is a new dataset + benchmark for predicting drug-target interactions, using Cell Painting data
Location: Fri 13 Dec 4:30 p.m. PST @ East Exhibit Hall A-C #4208
Poster: neurips.cc/virtual/2024...
Paper: arxiv.org/abs/2406.08649
Check out this work fresh off the (cell) press! We leveraged unbiased representations of cell morphology in high-content imaging for mechanistic understanding of inflammasome inhibitors. A similar approach can be applied to any mechanistic model with convergent pathways.
#ImmunoSky #microscopy
I don't think I saw this on social media at all: a bunch of very clever people (including our team) has put together a federated bioimage repository based on OME-Zarr with 500TB of data in 4 months! ome.github.io/ome2024-ngff...
Unfortunately most platforms use OTP via text as backup/fallback MFA (if not the only option), that is still ridiculously vulnerable (in the US).
uv is the life-saver you are looking for: github.com/astral-sh/uv
Perhaps the original CellPainting paper from the Broad folks is a good foundation for undergrads? Itβs a method paper, but lots of interesting research is built upon it.
www.nature.com/articles/npr...
Why flat out reject, though? Is it not possible to request the authors include license info?
