Graph Neural Networks and Biomedical Informatics

The field of biomedical informatics is moving towards increased adoption of graph neural networks (GNNs) to analyze complex biological data. Recent research has focused on developing frameworks and methods to improve the performance and interpretability of GNNs in this domain. A key direction is the integration of background knowledge (BK) into GNNs, which has shown promise in improving model performance. However, the impact of imperfect knowledge and the alignment of GNN architectures with BK characteristics are still being explored. Another area of research is the development of biological knowledge graphs, which can serve as a platform for advancing research in computational biology and precision medicine. Noteworthy papers include GNN-Suite, which provides a robust framework for constructing and benchmarking GNN architectures, and VitaGraph, which presents a comprehensive multi-purpose biological knowledge graph.TxPert is also notable for its use of biochemical relationships to improve out-of-distribution transcriptomic perturbation prediction.

Sources

GNN-Suite: a Graph Neural Network Benchmarking Framework for Biomedical Informatics

Informed, but Not Always Improved: Challenging the Benefit of Background Knowledge in GNNs

VitaGraph: Building a Knowledge Graph for Biologically Relevant Learning Tasks

Modeling Cell Dynamics and Interactions with Unbalanced Mean Field Schr\"odinger Bridge

Virtual Cells: Predict, Explain, Discover

TxPert: Leveraging Biochemical Relationships for Out-of-Distribution Transcriptomic Perturbation Prediction

Built with on top of