TRAILBLAZER: Generative Multicellular Perturbation Model of Biology
New Preprint on bioRxiv
We’re excited to share that our latest work, “TRAILBLAZER: generative multicellular perturbation model of biology,” is now available as a preprint on bioRxiv. This work introduces a new framework for modeling how biological systems respond to perturbations at the multicellular level while maintaining single-cell resolution.
Why This Matters
While recent single-cell foundation models have made significant progress, most approaches still treat cells as independent observations. In reality, tissues behave as coordinated systems, where interactions between cells define functional outcomes. This gap limits the ability of existing models to generalize across patients, conditions, and interventions.
What TRAILBLAZER Does
TRAILBLAZER addresses this challenge by modeling tissues as interacting systems using a multicellular transformer architecture. It combines a hyperspherical latent space with a generative decoder to enable zero-shot prediction of perturbation responses and ranking of candidate immunomodulators. By structuring latent space around shared biological references, the model supports meaningful extrapolation to unseen interventions.
Read the Full Preprint
Read the full preprint on bioRxiv
Read on bioRxiv →