New preprint with Garnett lab 🚀: from descriptive → causal single-cell atlases in CRC.We build the largest CRC atlas (>300 pts, 1.5M cells) using continual learning, and link cell states to causal drivers via Tahoe-100M, validated in organoids! www.biorxiv.org/content/10.6...

Check out the preprint for further details! The code is also available on GitHub.

Preprint: www.biorxiv.org/content/10.6...
Code: github.com/theislab/com...

Many thanks to our co-authors and collaborators
@gabrielepicco.bsky.social , @rasa-elmentaite.bsky.social, @teichlab.bsky.social, @elhamazizi.bsky.social, and others not on here for all their help along the way too! Grateful to collaborate across wonderful teams!

Thank you to @soroorzadeh.bsky.social whom I'm very grateful for all her brilliance and dedication in this project! I'm much appreciative to our supervisors, @mgarnett.bsky.social and @fabiantheis.bsky.social for their support and mentorship over the years. So great to work closely on this!

Together, we establish a framework to bridge observational atlases with causal perturbation biology, enabling the prediction and redirection of malignant cell fates.

In CRC organoids, we predict and validate that MAPK inhibition drives convergence of cells along an axis of proliferative and poor prognosis endoderm-like states, linking therapeutic perturbation to cell-state transitions and clinical outcome.

3. Causal mapping of perturbations to cell-state transitions - Using a reference-guided modelling framework, we quantify how pharmacological perturbations drive transitions across the phenotypic CRC landscape.

2. Clinically and genomically anchored malignant cell states - We identify hybrid endoderm-like states associated with KRAS mutation, MSS CRC, immune exhaustion, and poor clinical outcome, together with inflammatory states linked to MSI-high disease and favourable prognosis.

1. A new framework for comparative atlas construction - We introduce a continual learning strategy that preserves biological variation while enabling scalable integration, allowing us to resolve cellular states across the full continuum of CRC with unprecedented resolution.

Importantly, we embed this atlas within a shared phenotypic space with Tahoe-100M, directly linking cell states in CRC to causal determinants, which we validate in patient-derived organoids. Some highlights from our study:

We developed a continual learning framework that preserves disease trajectories and inter-patient heterogeneity to construct a comparative CRC atlas (>300 patients; 1.5 million cells).

Current colorectal cancer (CRC) atlases incompletely capture the malignant continuum from normal tissue to metastatic disease, and current integration approaches frequently obscure tumour-specific biology by over-correcting patient variation.

How do we address this?

New preprint from the labs of @mgarnett.bsky.social and @fabiantheis.bsky.social! A joint project with @soroorzadeh.bsky.social! 🧬🖥️ We show how to move from descriptive single-cell atlases to causal understanding of the mechanisms that define malignant cell states. biorxiv.org/content/10.6...

🧵👇

How can we hope to understand organismal development when it is controlled by huge, complex networks of interacting genes?


One option is to move away from molecular details and focus on learning representations and rules.

Check out the new perspective from me and @jamesbriscoe.bsky.social

Many congratulations to you and the team Maria! Very exciting work!

A Visual Guide to DNA Sequencing.

Learn how different DNA sequencing technologies work, from Sanger sequencing to Illumina to nanopores. (Complete with illustrations!)

Written by Evan DeTurk. Illustrated by Ella Watkins-Dulaney.

I haven’t spent enough time here, but this week Stripe Press launched our newest book, Maintenance: of Everything, Part One, by the inimitable Stewart Brand: press.stripe.com/maintenance-...

About the unglamorous yet civilizationally important work of maintenance and repair.

I have a good feeling I will! :) I've been enjoying titles from Stripe Press since the start.

I fondly go back a lot to the copy of "The Art of Doing Science and Engineering" that you gave me all those years ago. All the work you get up to at Stripe Press inspires me!

Big congrats on the launch!! My copy just arrived today and looking forward to diving in.

🌐Our new LMB website is live!🌐
Fresh design, better navigation, same first-class science.
Take a look around ➡️ mrclmb.ac.uk

new year, new job?!

Exciting postdoc position opening in my lab at the Wellcome Sanger Institute. Fully funded by CRUK

sanger.wd103.myworkdayjobs.com/en-US/Wellco...

find out more at www.coelho-lab.com

WRN inhibitors are being tested in patients with MSI cancers. How do tumours adapt to WRN inhibition, and can therapy stay ahead of resistance?

Our latest preprint investigates how MSI tumours evolve under pressure from WRN inhibition. Highlights below.

tinyurl.com/GenGenFaculty

GenGen is hiring group leaders! If you want to generate data at scale to train the next generation of AI models to make molecular biology predictive and programmable, we want to hear from you.

Generating long deletions across the genome with pooled paired prime editing screens Engineered deletions are a powerful probe for studying genome architecture, function, and regulation. Yet, the lack of effective methods to create them in large numbers and at multi-kilobase scale has...

New 🧬✂️ pre-print! We show that paired prime editing can efficiently generate large deletions — even >1 Mb — with high precision and at scale. We use this to perform the first pooled prime deletion screen across the human genome.

🔗 biorxiv.org/content/10.1...

A short thread (by Juliane Weller)👇

🚀 Excited to share SMART-PTA (miniaturized Primary Template Amplification) a high-throughput single-cell whole-genome amplification method that preserves clonal lineage information while enabling deep genomic and transcriptomic profiling.

Big, beautiful trees!!

SMART-PTA for whole-genome+transcriptome on thousand of single cells from the normal human esophagus 🤯 Massively scaling up the power of scWGS to build deep phylogenies and chart somatic evolution from birth throughout life.

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

GitHub - allydunham/dnacomb: CLI tool for flexibly parsing structured sequence reads into count tables and comparing them to expected libraries CLI tool for flexibly parsing structured sequence reads into count tables and comparing them to expected libraries - allydunham/dnacomb

We've been screening across many DNA construct structures with variable number/size of interesting regions - e.g. spacer, extension and barcode for prime editing. There wasn't a widely used tool to process these complex structured reads so I ended up developing github.com/allydunham/d....

Early career group leaders We appoint researchers from across biology and biomedicine to set up their first groups at the Crick.

The @crick.ac.uk is recruiting Early Career Group Leaders

- Lab set-up, research costs, salaries for up to 5 researchers
- Support for up to 12 years
- Access to our core facilities
- Competitive salary
- Fantastic colleagues
- All areas of biology

Deadline 27 Nov

www.crick.ac.uk/careers-stud...

In our latest paper, I wrote 2 supplementary notes (5 and 6) to briefly recap classical models, starting with Armitage&Doll and moving onto models with different types of clonal expansions. Some readers may find them of interest. [4/4] static-content.springer.com/esm/art%3A10...

A short thread. For years, I have been surprised by how much confusion our discovery of clones carrying cancer-driver mutations in normal tissues has caused in the cancer community. Typical questions like: (1) if you see these mutations in normal cells, are they really cancer drivers?... [1/4]