Related Experiment Video
Updated: Jun 22, 2025
07:52
Live Cell Imaging of Microtubule Cytoskeleton and Micromechanical Manipulation of the Arabidopsis Shoot Apical Meristem
Published on: May 23, 2020
5.3K
Graph metric learning quantifies morphological differences between two genotypes of shoot apical meristem cells in
Cory Braker Scott1,2,3, Eric Mjolsness2,3, Diane Oyen3
1Department of Mathematics and Computer Science, Colorado College, Colorado Springs, CO 80903, USA.
In Silico Plants
|June 28, 2024
Summary
This study introduces a novel graph learning method to analyze Arabidopsis thaliana cell division. The technique enhances graph diffusion distance (GDD) for improved classification and simulation analysis, revealing mutant cell characteristics.
Area of Science:
- Graph theory
- Computational biology
- Machine learning
Background:
- Graph diffusion distance (GDD) is a known measure for graph comparison.
- Analyzing biological processes like cell division often involves complex graph representations.
- Distinguishing between wild-type and mutant plant cells requires precise analytical methods.
Purpose of the Study:
- To develop a novel method for learning 'spectrally descriptive' edge weights in graphs.
- To generalize graph diffusion distance (GDD) for tunable loss minimization using neural networks.
- To apply this method for discriminating between wild-type and mutant Arabidopsis thaliana cell division graphs and analyzing cell division simulations.
Main Methods:
- Generalized graph diffusion distance (GDD) with differentiable steps for neural network optimization.
- Training edge weights and kernel parameters using contrastive loss for a learned distance metric.
- Application to graphs derived from Arabidopsis thaliana shoot apical meristem images (wild-type vs. trm678 mutants).
- Utilizing the trained model to compare biological graphs with simulated cell division graphs.
Main Results:
- Achieved a learned distance metric with large margins between graph categories (wild-type vs. mutant).
- Demonstrated improved performance of a k-nearest-neighbor classifier on the learned distance matrix.
- Identified simulation parameter regimes characterizing mutant versus wild-type Arabidopsis cells.
- Found trm678 mutant cells exhibit increased division plane randomness and reduced vertex avoidance.
Conclusions:
- The proposed method enables learning spectrally descriptive graph edge weights for enhanced discrimination.
- This approach successfully differentiates between wild-type and mutant Arabidopsis cell division patterns.
- The method provides insights into the cell division characteristics of trm678 mutants and aids in refining cell division simulations.
