Lightnews — Scholar-powered news

Paul Villoutreix

@paulvilloutreix.bsky.social

1.3K followers 700 following 79 posts

I develop machine learning frameworks to decode the geometry and topology of tissue development from spatial omics data | Junior Professor INSERM & Group Leader Turing Center for Living Systems | https://bioml.lis-lab.fr

bioml.lis-lab.fr

Posts Media Videos Starter Packs

Pinned

Paul Villoutreix @paulvilloutreix.bsky.social · Dec 31

Looking back, 2024 has been our most productive year. Here is a short thread on the various lines of work we've been braiding.

Our aim is to understand the relationship between single cell molecular states and tissue dynamics.

We develop methods to bridge transcriptomics and bioimage analysis

1 2 1

Reposted by Paul Villoutreix

Thomas Lecuit @thomaslecuit.bsky.social · 1d

Pleasure to announce the program of my new series of lectures at the College de France in Nov-Dec. Continuing on the theme of Biological information. I will focus on the computational aspects, building on David Marr's tri-level of analysis in biological systems. YouTube link sent when it starts.

1 6 19

Reposted by Paul Villoutreix

MOSAIC Group / Sbalzarini Lab @mosaicgroup.bsky.social · 7d

🤩👍🏼💪🏼 Meet #TopoSPAM - a high-performance simulation platform for biological morphogenesis with a friendly Python interface. @tudresden.bsky.social @csbdresden.bsky.social @mpi-cbg.de. Solve both continuous and discrete active mechanics! Released today: arxiv.org/abs/2509.24905

TopoSPAM: Topology grounded Simulation Platform for morphogenesis and biological Active Matter

We present a topology grounded, multiscale simulation platform for morphogenesis and biological active matter. Morphogenesis and biological active matter represent keystone problems in biology with ad...

arxiv.org

1 6 18

Reposted by Paul Villoutreix

Hallou Lab @halloulab.bsky.social · 19d

Very elegant and efficient methodology to recover spatial patterns from multi-modal spatial datasets 👌!! Many congratulations @paulvilloutreix.bsky.social and to all authors👏😀!!

1 3

Paul Villoutreix @paulvilloutreix.bsky.social · 19d

Link to the preprint ->
www.biorxiv.org/content/10.1...

jsPCA: fast, scalable, and interpretable identification of spatial domains and variable genes across multi-slice and multi-sample spatial transcriptomics data

Spatially structured cell heterogeneity within tissues is essential for healthy organ function. This heterogeneity is reflected by differential gene expression activity at various spatial location. Sp...

www.biorxiv.org

1 2

Paul Villoutreix @paulvilloutreix.bsky.social · 19d

jsPCA is simple, it is based on the product between the gene expression covariance (classical PCA) and the spatial autocorrelation.

We have shown that it's fast, interpretable and highly adaptable to multiple settings and large datasets.

Congrats to Ines Assali who led the project!

1 2

Paul Villoutreix @paulvilloutreix.bsky.social · 19d

We came up with an initial formulation of jsPCA last year when studying single cell morphometrics data
www.nature.com/articles/s41...
which was build on sPCA www.nature.com/articles/hdy....
To adapt it to spatial transcriptomics, we had to leverage sparsity and non convex optimization on manifold.

Single-cell morphometrics reveals T-box gene-dependent patterns of epithelial tension in the Second Heart field - Nature Communications

The embryonic heart tube undergoes elongation via the addition of progenitors from the second heart field, though how epithelial mechanics and genetics interact during this process remains unknown. He...

www.nature.com

Paul Villoutreix @paulvilloutreix.bsky.social · 19d

Finally, principal components of jsPCA are directly interpretable in terms of spatially variable genes (SVG). We found that the top 3000 genes of the first principal component recovered 80% of the SVGs obtained by the widely used SPARK-X method.

Paul Villoutreix @paulvilloutreix.bsky.social · 19d

jsPCA can also learn a joint representations on some of the slices and use it to predict the domains in an unseen slice.

Paul Villoutreix @paulvilloutreix.bsky.social · 19d

The joint analysis of multiple slices by jsPCA generates common domains among datasets, in contrast to monoslice analysis..

Paul Villoutreix @paulvilloutreix.bsky.social · 19d

jsPCA achieves state-of-the-art accuracy for domain identification while reducing computation from hours to seconds, and scales to atlas-level datasets such as Stereo-seq.

Paul Villoutreix @paulvilloutreix.bsky.social · 19d

Very happy to introduce jsPCA, a fast and interpretable computational framework for spatial transcriptomics that simultaneously identifies spatial domains and variable genes across multi-slice and multi-sample data.

2 8 17

Reposted by Paul Villoutreix

Aurelie Hintermann @aurhin.bsky.social · 21d

Are digits modified fins, or evolutionary innovations? Read how we tackled this old question from a new angle🧪
A story with @chasebolt.bsky.social, @homeobox.bsky.social and myself, coordinated by @denisduboule.bsky.social from @college-de-france.fr and published in @nature.com today!
#InHoxWeTrust

4 42 110

Paul Villoutreix @paulvilloutreix.bsky.social · 22d

This work was motivated by settings with scarce data such as rare diseases.

It turned into multi-dimensional puzzles and alignement problems that we really enjoyed solving! Congrats to Daniel Lepe-Soltero, the postdoc who led this work.

Link -> arxiv.org/abs/2503.18856

MODIS: Multi-Omics Data Integration for Small and unpaired datasets

An important objective in computational biology is the efficient integration of multi-omics data. The task of integration comes with challenges: multi-omics data are most often unpaired (requiring dia...

arxiv.org

1 5

Paul Villoutreix @paulvilloutreix.bsky.social · 22d

Finally, by performing targeted experiments, we showed that our approach was stable under severe class imbalance and (partially) missing class/modality pairs.

Paul Villoutreix @paulvilloutreix.bsky.social · 22d

We then applied our approach on the large public TCGA dataset and show that we could obtain high classification accuracy on up to 34 classes with three modalities

Paul Villoutreix @paulvilloutreix.bsky.social · 22d

We first show on simulated data that our architecture is stable under class imbalance, minimal supervision and partially missing class/modality pairs.

Paul Villoutreix @paulvilloutreix.bsky.social · 22d

To exploit labeled and unlabelled data, we proposed a semi-supervised approach, that enables latent space alignement.

Paul Villoutreix @paulvilloutreix.bsky.social · 22d

We learn multiple Variational Autoencoders with an auxiliary adversarial modality discriminator and a class discriminator operating in the latent space.
We made it easy to train and stable, by adapting the relativistic GAN loss and introduced multiple additional innovations to improve accuracy.

Paul Villoutreix @paulvilloutreix.bsky.social · 22d

Very happy to share MODIS - a new Multi-Omics Data Integration method for Small and unpaired datasets!

We solve a long-standing problem in the field: how to perform data integration for small unpaired datasets

Our solution: learning with class imbalance on both reference and target datasets!

1 2 13

Reposted by Paul Villoutreix

CNRS @cnrs.fr · 27d

#PressRelease 🗞️ The 2025 CNRS Gold Medal has been awarded to Stéphane Mallat, recognised the world over for his research on mathematics applied to signal processing and artificial intelligence.

👉 cnrs.fr/en/press/bet...

#CNRStalents 🏅

6 13

Reposted by Paul Villoutreix

Max-Planck-Gesellschaft @maxplanck.de · 29d

Springboard for an international scientific career! 🧬🧪🔭⚛️🧠🌱 Call for #MaxPlanckResearchGroups launched; applications are possible until October 14, 2025 www.mpg.de/max-planck-r... #ScienceCareer

2 56 52

Reposted by Paul Villoutreix

Janelia Research Campus @hhmijanelia.bsky.social · 29d

Become a Theory Fellow at Janelia.

🔍 Use computation & theory to tackle biological problems.
🖥️ Leverage state-of-the-art computational infrastructure
🤝 Collaborate across scientific disciplines

Apply by Nov. 3 @ https://janelia.link/theoryfellowprogram

11 19

Reposted by Paul Villoutreix

Ewan Birney @ewanbirney.bsky.social · Sep 8

Are you an AI expert who wants to stay in academia and change the world by understanding the most complex things we know - living organisms? Want to lead your own group, based in Heidelberg DE, working language English? @embl.org is hiring in AI embl.wd103.myworkdayjobs.com/en-US/EMBL/j...

Group Leader – AI in Biology

Are you ready to lead groundbreaking research in AI for Biology? Join us at EMBL! We are seeking a visionary scientist to establish their own independent research group bridging innovations in machine...

embl.wd103.myworkdayjobs.com

2 31 37

Reposted by Paul Villoutreix

EMBO Lab Leadership @embolableadership.bsky.social · Sep 5

Places available in our online courses this autumn: @embo.org Project Management (17-19.09), Negotiation for Scientists (18-19.09), Lab Leadership for Postdocs (07-09.10) and for Group Leaders (04-06.11) + Self-Leadership (14-16.10). Register now: www.embolableadership.org/our-courses/ #EMBOLabLead

People discussing projects at a pin board with text promoting upcoming online courses this autumn.

6 10

Reposted by Paul Villoutreix

Max Planck Institute of Molecular Cell Biology and Genetics @mpi-cbg.de · Sep 3

Dresden researchers @paveltomancak.bsky.social @bruvellu.bsky.social, Carl Modes, @cuencam15.bsky.social ‬ & colleagues published in @nature.com that a tissue fold in fruit fly embryos buffers mechanical stresses & may have evolved in response to mechanical forces. www.mpi-cbg.de/news-outreac...

Mechanical forces drive evolutionary change

A small tissue fold present in fruit fly embryos buffers mechanical stresses and may have evolved in response to mechanical forces.

www.mpi-cbg.de

3 30 81