Generalizable morphological profiling of cells by interpretable unsupervised learning.

Area of Science:

• Cell Biology
• Computational Biology
• Bioinformatics

Background:

• Traditional cell profiling is labor-intensive and biased.
• Deep learning offers alternatives but lacks interpretability and requires labeled data.

Purpose of the Study:

• Introduce MorphoGenie, an unsupervised deep learning framework for single-cell morphological profiling.
• Develop a method that overcomes the curse of dimensionality and provides interpretable results.

Main Methods:

• Utilized disentangled representation learning and high-fidelity image reconstruction.
• Created a compact, interpretable latent space for capturing biologically meaningful features.
• Linked latent representations to hierarchical morphological attributes for semantic interpretability.

Main Results:

• Achieved robust performance across diverse imaging modalities and experimental conditions.
• Enabled accurate cell type/state classification and continuous trajectory inference.
• Revealed cellular behaviors often missed by manual examination.

Conclusions:

• MorphoGenie provides a generalized, unbiased strategy for morphological profiling.
• The framework enhances the quantitative and data-driven transformation of cell biology.
• Offers a powerful tool for discovering novel cellular insights.