Reproduce “GRAPE: Heterogeneous Graph Representation Learning for Genetic Perturbation with Coding and Non-Coding Biotype” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Changxi Chi, Jun Xia, Jingbo Zhou, Jiabei Cheng, Chang Yu, Stan Z. Li

GRAPE, a heterogeneous graph neural network, integrates multi-modal gene features and biotype-aware graph structure learning to accurately predict genetic perturbation effects in single cells. The model achieves state-of-the-art performance, significantly improving RMSE by approximately 7% and Direction Accuracy Coefficient (DAC) by 6% over prior methods on benchmark datasets.

Key insights

  • Integrating multi-modal biological data (text and DNA sequences) provides a richer, more comprehensive gene representation than single-modality approaches, crucial for accurate prediction.
  • Dynamically learning and refining the gene regulatory network structure (via K-hop GSL) is essential for uncovering complex, hidden gene interactions and multi-scale regulatory dynamics.
  • Explicitly modeling gene biotypes (coding vs. non-coding) through a heterogeneous graph and HGAT allows the model to capture distinct functional roles and interactions, significantly enhancing predictive power and learning biologically meaningful gene embeddings.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.