The paper presents an overview of Tahoe-x1 (Tx1), a new family of perturbation-trained single-cell foundation models developed by Tahoe Therapeutics, with up to three billion parameters. The authors explain that foundation models, which have transformed fields like computer vision, are now being adapted for single-cell biology, particularly for applications in precision oncology. Tx1 is pretrained on massive datasets, including the Tahoe-100M perturbation compendium, to learn generalized representations of genes, cells, and compounds. The model achieves state-of-the-art performance across multiple cancer-relevant benchmarks, such as predicting gene essentiality and modeling cellular responses to various chemical perturbations. Ultimately, the research emphasizes that scaling model capacity and training on large-scale perturbation data are crucial for developing robust, transferable models for cancer therapeutics.

References:

• Gandhi S, Javadi F, Svensson V, et al. Tahoe-x1: Scaling Perturbation-Trained Single-Cell Foundation Models to 3 Billion Parameters[J]. bioRxiv, 2025: 2025.10. 23.683759.