Deep-learning model predicts how fruit flies form, cell by cell
20 hours ago
- #deep learning
- #cell dynamics
- #developmental biology
- MIT engineers developed a deep-learning model to predict minute-by-minute cell behavior during fruit fly embryo development.
- The model tracks geometric properties of cells, such as position and neighbor contact, with 90% accuracy in predictions.
- Potential applications include predicting development in more complex organisms and identifying early disease patterns like asthma and cancer.
- The model uses a 'dual-graph' structure, combining point cloud and foam approaches to represent cell dynamics.
- Trained on high-quality videos of fruit fly gastrulation, the model predicts cell folding, division, and neighbor interactions.
- Future goals include applying the model to other species and human tissues, pending availability of high-quality data.