Morphology-robust quantification of subcellular organization in complex cells.

Area of Science:

• Cell Biology
• Computational Biology
• Microscopy Image Analysis

Background:

• Quantitative analysis of subcellular protein organization is challenging due to cell morphology variations in complex cells.
• Interpreting protein localization patterns in fluorescence microscopy data is often confounded by these morphological differences.

Purpose of the Study:

• To introduce CellAligner, an unsupervised framework for morphology-robust comparison of subcellular localization.
• To enable accurate analysis of protein organization in morphologically diverse cell types, such as neurons and glia.

Main Methods:

• Utilized fused unbalanced Gromov-Wasserstein couplings to map protein distributions from distinct cells into shared anchor-cell geometries.
• Developed dCellAligner-OT, a deep metric learning model for scalable, atlas-level analyses by approximating optimal transport distances.

Main Results:

• CellAligner significantly reduced morphology-associated confounding in neuronal imaging benchmarks.
• Improved multiclass MCC for localization classification by approximately twofold when applied with existing image-analysis methods.
• Identified U18666A-induced lysosomal trafficking defects in human iPSC-derived neurons, demonstrating biological utility.

Conclusions:

• CellAligner provides a generalizable framework for robust analysis of subcellular organization in complex cellular systems.
• The approach enhances the accuracy and interpretability of fluorescence microscopy data from morphologically varied cells.