Multi-modal tissue-aware graph neural network for in silico genetic discovery

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Understanding tissue-specific gene function is crucial for identifying drug targets.
• Current computational methods often overlook context-dependent gene interactions.

Purpose of the Study:

• Introduce Mahi, a graph neural network framework for learning context-aware gene representations.
• Integrate multi-modal molecular features within tissue-specific contexts.
• Improve prediction of gene essentiality and perturbation responses.

Main Methods:

• Developed a scalable and interpretable graph neural network (GNN) framework named Mahi.
• Integrated chromatin accessibility, transcription factor binding, histone modifications, and protein structure data.
• Pretrained on tissue-specific network topologies and then integrated multi-modal features across 290 tissues/cell-types.

Main Results:

• Mahi learns context-aware gene embeddings, outperforming sequence-based models in predicting gene essentiality across 1,183 cancer cell lines.
• The learned embedding space reveals tissue-specific gene organization and context-dependent roles.
• Simulated gene knockout perturbations using Mahi identified disease-relevant pathways and potential therapeutic targets.

Conclusions:

• Mahi provides a foundation for modeling tissue-specific gene function and perturbation responses.
• The framework enables applications in precision medicine, drug discovery, and identifying context-dependent genetic vulnerabilities.
• All learned embeddings and the Mahi framework are publicly available.