OCOO-T : A Simple and Scalable Virtual Cell Model for Transcriptional Perturbation Response Prediction

2026-06-12 · Source: cs.AI updates on arXiv.org · Field: Science & Research — Life Sciences & Biology, Mathematics & Computational Sciences, Artificial Intelligence & Machine Learning · Depth: Expert, extended

Summary

OCOO-T is a minimalist flow-matching-based AI Virtual Cell (AIVC) model designed for predicting single-cell transcriptional responses to genetic, chemical, and cytokine perturbations. Developed by INFevo, it employs a vanilla Transformer stack that directly processes continuous gene expression profiles, framing prediction as a continuous-time denoising process. Perturbation embeddings, dosage, and cell-line/cell-type specificity are integrated via adaptive layer normalization and in-context tokens. Evaluated on Tahoe100M, Replogle, and PBMC benchmarks, OCOO-T achieves strong performance across diverse perturbations and cell types. It scales to long transcriptional profiles (e.g., 18,080 genes) through a patching and depatching mechanism, demonstrating that architectural simplicity can yield competitive results without complex auxiliary modules or latent spaces.

Key takeaway

For AI Scientists and Machine Learning Engineers developing virtual cell models, OCOO-T demonstrates that focusing on a minimalist Transformer architecture with direct gene expression processing and flow-matching denoising can achieve strong results. You should consider simplifying your model designs, leveraging patching for full-transcriptome analysis, and exploring direct covariate conditioning to enhance scalability and generalizability in perturbation response prediction. This approach avoids complex latent spaces and auxiliary encoders, streamlining development.

Key insights

OCOO-T uses a simple Transformer and flow matching to directly denoise gene expression for perturbation prediction.

Principles

• Directly model continuous gene expression.
• Architectural simplicity is sufficient.
• Patching enables long sequence processing.

Method

OCOO-T formulates prediction as continuous-time denoising using Rectified Flow. It interpolates between Gaussian noise and target response, training a Transformer to predict velocity. Inference integrates this velocity field via an ODE solver.

In practice

• Use ESM2 for gene identity embeddings.
• Employ AdaLN for condition injection.
• Consider mean control-cell profiles for unseen contexts.

Topics

• Virtual Cell Models
• Transcriptional Perturbation Prediction
• Flow Matching
• Transformer Architectures
• Single-cell Omics
• Gene Expression Denoising

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Discovery of candidate therapeutic targets with Geneformer — Geneformer leverages a transformer model and transfer learning for gene network analysis and therapeutic target discovery with limited data.
• Machine learning framework reveals a concordant cell-state landscape across single-cell datasets — CONCORD is a contrastive learning framework for integrating single-cell omics data to reveal cell-state landscapes.
• [R] Genomic Large Language Models — Genomic LLMs can identify functional similarities in DNA beyond sequence alignment, indicating learned regulatory patterns.
• Identifying spatial single-cell-level interactions with graph transformer — A self-supervised graph transformer identifies spatial single-cell interactions without known ligand-receptor pairs.
• $\textit{BlockFormer}$ : Transformer-based inference from interaction maps — BlockFormer infers per-entity parameters from variable interaction maps using synthetic data and a block-aware transformer with 3D positional encoding.
• Knowledge Graphs and Reasoning LLMs for Finding Simple Yet Effective Transcriptomic Perturbation Predictors — Simple K-nearest neighbour models leveraging knowledge graphs, enhanced by RL-trained LLMs, offer competitive transcriptomic perturbation prediction.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.