Anomaly detection for high-content image-based phenotypic cell profiling.

Area of Science:

• Cell Biology
• Bioinformatics
• Computational Biology

Background:

• High-content image-based phenotypic profiling uses automated microscopy to analyze cell morphology and infer physiological states.
• Classical phenotypic profiles struggle to capture complex cellular organization, and current machine learning methods lack biological interpretability.
• Developing interpretable and comprehensive cell representation methods is crucial for advancing biological insights.

Purpose of the Study:

• To develop a novel self-supervised anomaly-based representation for high-content phenotypic profiling.
• To enhance the interpretability and capture of complex morphological inter-feature dependencies in cell representations.
• To evaluate the performance of anomaly-based representations against classical methods in downstream biological tasks.

Main Methods:

• Utilized control wells to define the in-distribution of experimental data.
• Formulated a self-supervised reconstruction anomaly-based representation method.
• Evaluated anomaly-based representations on four public Cell Painting datasets for reproducibility and mechanism of action classification.
• Employed unsupervised explainability techniques for autoencoder-based anomalies.

Main Results:

• Anomaly-based representations demonstrated improved reproducibility in phenotypic profiling.
• Enhanced classification accuracy for mechanism of action prediction compared to classical representations.
• Successfully encoded intricate morphological inter-feature dependencies while maintaining interpretability.
• Unsupervised explainability identified specific inter-feature relationships driving anomalies.

Conclusions:

• Anomaly-based representations offer a powerful and interpretable approach for high-content image-based phenotypic profiling.
• This method complements existing techniques and improves key downstream biological applications.
• The anomaly-based representation concept is adaptable to various cell biology research areas.