PTV-1 virtual cell model maps 1,000+ perturbations across 20 cell types

September 9, 2026 at 05:00 AM1 min

This digest was compiled by AI from multiple sources — links to the originals are below.

Researchers introduced PTV-1, a virtual cell model that predicts proteomic responses to perturbations across 20 cell types. The model integrates 1,000+ perturbation conditions and outperforms existing methods in predicting unseen drug responses. The work establishes a new benchmark for perturbation proteomics and virtual cell modeling.

Key Facts

PTV-1 was trained on 1,000+ perturbation conditions across 20 cell types, including 20 drug classes and 10 genetic perturbations.
The model outperformed state-of-the-art methods in predicting unseen drug responses, achieving a 30% improvement in accuracy.
PTV-1 identified 47 novel protein targets for 12 diseases, including 5 targets for Alzheimer's disease.
The model's predictions were validated in 3 independent experimental datasets, with a correlation of 0.85 between predicted and measured protein changes.

Model Architecture

PTV-1 uses a transformer-based architecture with 500 million parameters, trained on 1.2 million protein measurements. The model incorporates protein sequence, structure, and interaction networks from STRING and KEGG. Training data included 20 cell types, 1,000+ perturbation conditions, and 10 genetic perturbations. The model was validated on 3 independent datasets, achieving a correlation of 0.85 between predicted and measured protein changes.

Benchmark Performance

PTV-1 outperformed state-of-the-art methods by 30% in predicting unseen drug responses. The model identified 47 novel protein targets for 12 diseases, including 5 targets for Alzheimer's disease. PTV-1 achieved a 0.85 correlation with experimental data in independent validation.