Predicting mitochondrial fission, fusion and depolarisation event locations from a single z-stack

Area of Science:

• Cell Biology
• Artificial Intelligence
• Biophysics

Background:

• Mitochondrial dynamics, including fission, fusion, and depolarization, are crucial for cellular health.
• Traditional methods for studying these events require time-lapse imaging, which can be complex and time-consuming.

Purpose of the Study:

• To develop a novel method for predicting mitochondrial fission, fusion, and depolarization events in 3D using only single-cell morphology.
• To assess the feasibility of using generative adversarial networks (GANs) for this predictive task.

Main Methods:

• Implementation of a 3D Pix2Pix GAN and a 3D adversarial segmentation network (Vox2Vox GAN).
• Training and evaluation of the networks using morphological data from mitochondria.

Main Results:

• The Pix2Pix GAN achieved prediction accuracies of 35.9% (fission), 33.2% (fusion), and 4.90% (depolarization).
• The Vox2Vox GAN achieved prediction accuracies of 37.1% (fission), 37.3% (fusion), and 7.43% (depolarization).
• While current accuracies are too low for direct application, the networks show promise in modeling mitochondrial dynamics.

Conclusions:

• The developed AI models demonstrate a novel approach to predicting mitochondrial events from static images.
• These findings establish a baseline for future research in AI-driven mitochondrial dynamics analysis.
• The method has the potential to streamline research and drug trials by reducing the need for time-lapse imaging.