iiinova — where your digital self keeps evolving

Zev Gartner

@zevgartner.bsky.social

396 followers 167 following 23 posts

Building tissues to understand how tissues build themselves

Posts Replies Media

Zev Gartner @zevgartner.bsky.social · 13/01/2026

6. It's not just for scRNA-seq. CONCORD seamlessly handles and aligns scATAC-seq and spatial transcriptomics (Xenium, etc.), revealing fine-grained subtypes across technologies

100

Zev Gartner @zevgartner.bsky.social · 13/01/2026

5. CONCORD resolved complex topologies during intestinal development, such as differentiation trajectories linked to cell cycle loops and providing biologically interpretable encodings traced to gene co-expression programs (click here for interactive 3D UMAP! qinzhu.github.io/Concord_docu...):

120

Zev Gartner @zevgartner.bsky.social · 13/01/2026

4. On a whole-organism, multi-species embryonic atlas of C. elegans and briggsae from @jisaacmurray’s lab where lineage annotations provide a ground truth, CONCORD resolved lineage trajectories and bifurcations (see animation and click here for interactive 3D UMAP! qinzhu.github.io/Concord_docu...):

110

Zev Gartner @zevgartner.bsky.social · 13/01/2026

3. With this sampling scheme, we found that even a minimalist single-layer network achieves better-than-SOTA performance. It disentangles biology from batch effects without over-correcting. We benchmarked it rigorously. Across a diverse set of datasets, CONCORD consistently ranked top: ben

130

Zev Gartner @zevgartner.bsky.social · 13/01/2026

2. At the core is a novel mini-batch sampler that changes how the model 'sees' data. Unlike uniform sampling, our approach is two-fold: 🔹 Dataset-aware: Extracts consistent biological structure across batches. 🔹 Hard-negative: Distinguishes differences between closely related cell states.

100

Zev Gartner @zevgartner.bsky.social · 13/01/2026

1. Excited to share CONCORD, out in @NatureBiotech, an ML framework for single-cell analysis addressing integration, dimensionality reduction, and denoising in one go by @qinzhu1.🔗 www.nature.com/articles/s41... Check out this CONCORD model of worm development resolving differentiation trajectories:

1134

Zev Gartner @zevgartner.bsky.social · 25/10/2025

Mind blown 🤯

000

Zev Gartner @zevgartner.bsky.social · 20/03/2025

7. The key innovation leading to this performance gain is a probabilistic, dataset- and neighborhood-aware sampling strategy tailored for self-supervised contrastive learning, enabling the generation of coherent, denoised and high-resolution latent cell representations across diverse datasets:

neighborhood and dataset aware sampling improves learning

120

Zev Gartner @zevgartner.bsky.social · 20/03/2025

5. An intestinal development atlas demonstrated CONCORD's ability to resolve complex topologies, including differentiation trajectories intertwined with cell cycle loops (click here for interactive 3D UMAP! qinzhu.github.io/Concord_docu...):

loops, cylinders and filaments in a gut developmental atlas

110

Zev Gartner @zevgartner.bsky.social · 20/03/2025

4. We validated CONCORD on an embryonic atlas of C. elegans and C. briggsae from @jisaacmurray. CONCORD captured known bifurcations, resolved subtle cell states, and uncovered lineage convergence events missed by other methods (click here for interactive 3D UMAP! qinzhu.github.io/Concord_docu...):

see lineage relationships in C. elegans developmental data sets

120

Zev Gartner @zevgartner.bsky.social · 20/03/2025

3. We evaluated CONCORD on datasets with topologies such as continuous trajectories, loops, clusters and hierarchical trees. Using scIB metrics, geometric measures, and TDA, we assessed its ability to preserve biological structure. CONCORD outperforms existing methods:

CONCORD outperforms other methods on metrics of geometry and topology

100

Zev Gartner @zevgartner.bsky.social · 20/03/2025

1. Struggling to integrate single-cell datasets? Finding it hard to resolve clear differentiation trajectories? Reveal the underlying structure in your data with CONCORD.

1226