Author | Lightnews

Tina Šubic @tincamalinca.bsky.social · 11d

Thanks! Happy to hear any feedback 🙂

Tina Šubic @tincamalinca.bsky.social · 11d

I had the best time working on dLEM! Huge thanks to my amazing co-authors: Maria Chikina, Ali Tugrul Balci, Kristina Perevoshchikova, and @gfudenberg.bsky.social!

check the preprint: 🔗 www.biorxiv.org/content/10.1...

Mechanistic Genome Folding at Scale through the Differentiable Loop Extrusion Model

The spatial folding of the genome shapes gene regulation by controlling which loci interact, yet inferring the mechanisms behind these 3D structures from contact maps remains difficult. Cohesin-mediat...

www.biorxiv.org

Tina Šubic @tincamalinca.bsky.social · 11d

Bottom line: dLEM shows we don't have to choose between mechanistic insight and predictive power.
By reformulating loop extrusion as a differentiable process, we get:
✅ Interpretability
✅ Scalability
✅ Perturbation prediction
✅ Integration with ML

10/10

1

Tina Šubic @tincamalinca.bsky.social · 11d

The parameter reduction compared to other deep learning models in deep dLEM is wild 🤯:
- 280× fewer than C.Origami
- 750× fewer than Orca
Yet we achieve competitive accuracy!

Why? Because we hard-coded the biophysics instead of making the model learn it.

9/10

1

Tina Šubic @tincamalinca.bsky.social · 11d

We took dLEM to the next level: deep dLEM 🧠
We embedded dLEM as a biophysical layer in a neural network that learns from:

- DNA sequence
- Chromatin accessibility (DNase/ATAC)

Result: Sequence → structure prediction that stays mechanistically grounded

8/10

1

Tina Šubic @tincamalinca.bsky.social · 11d

Example 2: CTCF depletion
→ We systematically reduced CTCF coefficients in our linear model
→ Optimal prediction at α=0.33, matching the 33% of CTCF peaks that remain after auxin treatment
The model reveals residual CTCF is still functionally active!

7/10

1

Tina Šubic @tincamalinca.bsky.social · 11d

Because dLEM parameters are physically interpretable, we can do something deep learning can't: predict perturbation effects!

Example: WAPL depletion
→ We modified just the detachment rate
→ Successfully predicted emergence of new loops & extended stripes

1

Tina Šubic @tincamalinca.bsky.social · 11d

What do we learn from these velocity profiles?
😌 CTCF shows expected directional asymmetry (blocks L or R depending on orientation)
😲 BUT: Transcription machinery (H3K4me3, H3K36me3, H3K4me1,...) ALSO modulates cohesin dynamics!
CTCF isn't the whole story 👀

5/10

1

Tina Šubic @tincamalinca.bsky.social · 11d

dLEM acts as an encoder-decoder.
📥 ENCODER: Compress complex 2D Hi-C maps into simple 1D velocity profiles
📤 DECODER: Reconstruct contact maps from these profiles

It's dimensionality reduction, but with built-in biophysics!

4/10

1

Tina Šubic @tincamalinca.bsky.social · 11d

💡The key insight: While contact maps are 2D, loop extrusion is inherently 1D—cohesin walks along chromatin!
dLEM captures this with two velocity profiles:
→ L: leftward cohesin speed
→ R: rightward cohesin speed
Obstacles (like CTCF) = dips in velocity 📉

3/10

1

Tina Šubic @tincamalinca.bsky.social · 11d

The challenge: Current methods face an impossible tradeoff.
❌ Polymer simulations = mechanistic, but not scalable
❌ Deep learning = predictive, but black-box
❌ Summary metrics = compressed, but mechanism-agnostic
We needed something that's mechanistic, interpretable, AND scalable.
✅ Enter dLEM!
2/10

1

Tina Šubic @tincamalinca.bsky.social · 11d

✨New preprint!✨

We built dLEM - a differentiable Loop Extrusion Model that bridges biophysics and machine learning for 3D genome folding.

dLEM makes loop extrusion learnable and interpretable—predicting how genomes fold and how they respond to perturbations.

🧵

bioRxiv Genomics @biorxiv-genomic.bsky.social · 15d

Mechanistic Genome Folding at Scale through the Differentiable Loop Extrusion Model https://www.biorxiv.org/content/10.1101/2025.10.17.682904v1

2 3 4

Reposted by Tina Šubic

MOSAIC Group / Sbalzarini Lab @mosaicgroup.bsky.social · Dec 21

Learning interpretable mathematical models from data is transforming science. But what if different models apply in different areas or at differnt times? After all, living systems are plastic and self-organized. Now you can learn *local* model clusters! doi.org/10.1098/rspa...

Learning locally dominant force balances in active particle systems | Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences

We use a combination of unsupervised clustering and sparsity-promoting inference algorithms to learn locally dominant force balances that explain macroscopic pattern formation in self-organized active...

doi.org

1 2