Image-derived, three-dimensional generative models of cellular organization.

Area of Science:

• Cellular Biology
• Computational Biology
• Statistical Learning

Background:

• Protein subcellular location is crucial for function and computational cell modeling.
• Previous 2D models captured subcellular patterns and cell-to-cell variations from microscope images.
• These models included sub-models for nuclear/cell shape and organelle distribution.

Purpose of the Study:

• To extend 2D generative models to three dimensions for more accurate protein subcellular location modeling.
• To enable the synthesis of multi-channel images with multiple proteins, overcoming limitations of direct microscopy.
• To provide a compact and interpretable representation of subcellular patterns for analyzing biological changes.

Main Methods:

• Developed a 3D generative model framework based on statistical learning.
• Extended previous models to incorporate three-dimensional cellular structures.
• Enabled combination of different pattern models for multi-protein simulations.

Main Results:

• Achieved more accurate descriptions of protein subcellular locations in 3D.
• Facilitated the creation of synthetic multi-channel images representing multiple proteins.
• Model parameters offer a concise and interpretable way to describe subcellular organization.

Conclusions:

• The 3D generative model significantly enhances the simulation of protein distributions within cells.
• This approach facilitates the study of subcellular organization and its alterations.
• The model holds potential for automated identification of changes induced by perturbagens.