Interpretable representation learning for 3D multi-piece intracellular structures using point clouds.

Area of Science:

• Cell Biology
• Computational Biology
• Bioinformatics

Background:

• Quantifying intracellular structures with complex morphologies is challenging.
• Objective, robust, and generalizable measurements are needed for understanding subcellular organization.

Purpose of the Study:

• To develop a morphology-appropriate representation learning framework for complex intracellular structures.
• To enable objective and interpretable quantification of cellular components.

Main Methods:

• Utilized three-dimensional rotation-invariant autoencoders and point clouds for representation learning.
• Applied the framework to punctate (e.g., DNA replication foci) and polymorphic (e.g., nucleoli) intracellular structures.
• Benchmarked performance against image-based autoencoders using efficiency, generative capability, and representation expressivity metrics.

Main Results:

• The framework learns orientation-independent, compact, and interpretable shape representations.
• Demonstrated unsupervised discovery of subclusters within complex structures.
• Successfully applied the approach to phenotypic profiling of nucleoli after drug perturbations.

Conclusions:

• The proposed framework effectively quantifies complex intracellular structures by embracing their morphology.
• This approach facilitates unsupervised discovery and aids in phenotypic analysis.
• The method offers a robust tool for advancing the study of subcellular organization.