Related Experiment Video
Updated: Aug 5, 2026
13:34
A Combined 3D Tissue Engineered In Vitro/In Silico Lung Tumor Model for Predicting Drug Effectiveness in Specific Mutational Backgrounds
Published on: April 6, 2016
A Controlled in Silico Benchmark for GNN Prediction of Tissue Dynamics
Matej Krajnc1, Troy Comi2, Siqi Miao3
1Department of Theoretical Physics, Jožef Stefan Institute, Ljubljana, Slovenia.
Research Square
|July 29, 2026
Summary
This study introduces an in silico benchmark for Graph Neural Networks (GNNs) to predict tissue dynamics. The benchmark reveals that specific GNN architectures and input features significantly impact prediction accuracy, especially in disordered tissues.
Area of Science:
- Computational Biology
- Biophysics
- Machine Learning
Background:
- Graph Neural Networks (GNNs) show promise for predicting tissue dynamics.
- Selecting optimal GNN architectures is challenging due to limited, noisy, and system-specific experimental data.
Purpose of the Study:
- To create a controlled in silico benchmark for comparing GNN architectures in predicting tissue dynamics.
- To evaluate GNN performance on a vertex-model task: predicting cell-cell interface lengths after neighbor exchange.
Main Methods:
- Developed a simulated tissue environment for controlled GNN benchmarking.
- Varied tissue geometry, mechanics, perturbation complexity, dataset size, and input features.
- Assessed performance of Provably Powerful Graph Networks (PPGN) and Principal Neighborhood Aggregation (PNA) architectures.
Main Results:
- PPGN and PNA were most sample-efficient when provided with pre-event edge lengths.
- Performance decreased significantly when only topology was used as input.
- A 'predict-or-copy' strategy was observed, where distant predictions reverted to pre-event lengths.
- Prediction difficulty correlated with tissue disorder, indicated by hexagonality.
Conclusions:
- The in silico benchmark provides a reproducible method for evaluating GNNs in tissue remodeling.
- Identified key factors influencing GNN performance, including input features and tissue geometry.
- The findings aid in diagnosing feature dependence, copying behavior, and geometric consistency in GNN predictions for biological systems.
