X-Pert Documentation

X-Pert is a transformer-based framework that unifies genetic, chemical, and combinatorial perturbation modeling. It couples a Perturbation Perceiver—which embeds heterogeneous perturbations into a shared latent Perturbverse—with a Cell Encoder that fuses gene identity, expression, and perturbation-impact tokens. This architecture captures hierarchical gene–perturbation and gene–gene dependencies, enabling accurate predictions for unseen, dose/efficacy-aware, and combinatorial perturbations while supporting downstream analyses such as perturbation retrieval and drug–gene association discovery.

X-Pert model overview

X-Pert couples a Perturbation Perceiver with a Cell Encoder to model diverse perturbation responses.

News

  • 2025-11-12 — X-Pert officially goes open source on GitHub, sharing the full in silico perturbation workflow with the community.

Key Capabilities

  • Unified Perturbation Space – Align genetic and chemical perturbations in a shared latent representation for cross-type analysis and retrieval.

  • Hierarchical Response Modeling – Combine gated cross-attention and self-attention to respect perturbation-specific regulatory cascades and pathway programs.

  • Dose & Efficacy Awareness – Incorporate quantitative perturbation strength to improve predictions across variable dosage and sgRNA efficacy.

  • Scalable Benchmarks – Achieve strong performance across single-cell and bulk datasets, including unseen perturbations, combinations, and large-scale screens.

  • Downstream Discovery – Support perturbation retrieval, drug repurposing, and interpretable embedding analyses within the Perturbverse.

Getting Started

Tutorial Notebooks

Resources

Indices and Tables