Related Experiment Video
Updated: May 9, 2026

Generation and Downstream Analysis of Single-Cell and Single-Nuclei Transcriptomes in Brain Organoids
Published on: March 29, 2024
TranscriptFormer: A generative cell atlas across 1.5 billion years of evolution.
James D Pearce1, Sara E Simmonds1, Gita Mahmoudabadi1
1Biohub, Redwood City, CA, USA.
We created TranscriptFormer, a powerful AI model that analyzes single-cell data across species and evolutionary time. It reveals universal cellular principles and accurately classifies cell types, even across vast evolutionary distances.
Area of Science:
- Computational Biology
- Evolutionary Biology
- Genomics
Background:
- Single-cell transcriptomics offers insights into cellular diversity.
- Comparing transcriptional data across species and evolutionary history is complex.
- Existing methods struggle with deep evolutionary comparisons.
Purpose of the Study:
- To develop a scalable computational framework for cross-species single-cell transcriptomic analysis.
- To leverage foundation models for understanding evolutionary principles of cellular organization.
- To enable accurate cell type classification and disease state identification across diverse species.
Main Methods:
- Development of TranscriptFormer, a family of generative foundation models.
- Training on a large dataset of 112 million cells from 12 species, covering 1.53 billion years of evolution.
- Evaluating performance on cell type classification, zero-shot disease state identification, and emergent biological insights.
Main Results:
- TranscriptFormer achieves state-of-the-art cell type classification accuracy, even for species diverged over 685 million years.
- The model demonstrates zero-shot disease state identification in human cells.
- Developmental trajectories, phylogenetic relationships, and cellular hierarchies are learned implicitly by the model.
- Universal principles of cellular organization are identified and predictable across the tree of life.
Conclusions:
- TranscriptFormer provides a powerful framework for quantitative single-cell analysis and comparative cellular biology.
- Foundation models can learn and predict universal biological principles from large-scale transcriptomic data.
- This approach opens new avenues for understanding evolution and cellular function across diverse life forms.
Related Concept Videos
Gene Evolution - Fast or Slow?
In contrast, regions which code...
Gene Evolution - Fast or Slow?
In contrast, regions which code...
Genome Size and the Evolution of New Genes
Genome Size and the Evolution of New Genes
Cell Specific Gene Expression
The Tree of Life - Bacteria, Archaea, and Eukaryotes

