work with @jmtomczak.bsky.social sudarshan babu Payam Dibaeinia Donghui Li, PhD Aly A. Khan @tkaraletsos.bsky.social
Yesterday, we released a new paper on a novel architecture for single-cell genomics. We call it scLDM to hint at the Latent Diffusion Model paradigm, a generative modeling framework that has been very successful across various domains.
I'll be at the AI Virtual Cells and Instruments: A New Era in Drug Discovery and Development.
Stop by to chat about generative models and virtual cells! I'll be presenting our work @biohub.org on a scalable latent diffusion model for single-cell gene expression data.
We are excited because it presents a novel way to learn "latent tokens" of gene expression variation, overcoming many of the limitations of order(less) tokenization and learning schemes. We show its performance in generative and representation learning tasks.
Giovanni Palla
For further details, check out:
📜 our paper: arxiv.org/abs/2511.02986
💻 our code: github.com/czi-ai/scldm
🖥️ a blogpost: biohub.org/blog/scldm-a...
🤖 a model card: virtualcellmodels.cziscience.com/model/scldm
As more single-cell perturbation data become available, it is crucial to establish a consensus on measures of success for perturbation prediction models.
I reviewed recent literature and collected some thoughts on the current state of affairs.
open.substack.com/pub/giovanni...
Fantastic lineup of panelists to discuss everything agentic coding in bioinformatics, still time to register here:
www.eventbrite.com/e/scverse-co...
Fantastic collaboration with @jmtomczak.bsky.social , the Biohub Chicago team sudarshan babu Payam Dibaeinia Aly A. Kahn and the team in the Bay Area James Douglas Pearce Donghui Li, PhD Theofanis Karaletsos.
Giovanni Palla
Giovanni Palla
Giovanni Palla
Happy to share CellAnnotator, a lightweight Python package to query OpenAI models for initial cell type annotations: github.com/quadbio/cell...
Inspired by ideas in www.nature.com/articles/s41... and github.com/VPetukhov/GP..., implemented as an scVerse ecosystem package with docs and tutorials.
Thoughts on metrics for perturbation prediction models
Excited to see a home for MCP servers in bioinformatics emerge! The community needs a central hub for developing MCP servers for key bioinfo tools. At @scverse.bsky.social we're contributing to this effort - can't wait to share what we're building
