Capturing cell heterogeneity in representations of cell populations for image-based profiling using contrastive...

Area of Science:

• Computational biology
• Cellular imaging
• Machine learning

Background:

• Image-based cell profiling measures thousands of single-cell features to compare perturbed cell populations.
• Averaging single-cell profiles masks crucial cellular heterogeneity, limiting predictive accuracy.

Purpose of the Study:

• To introduce CytoSummaryNet, a novel Deep Sets-based method for improved aggregation of single-cell feature data.
• To enhance mechanism of action prediction in image-based cell profiling by capturing cellular heterogeneity.

Main Methods:

• CytoSummaryNet employs self-supervised contrastive learning within a multiple-instance learning framework.
• The approach aggregates single-cell profiles, prioritizing informative cells and downweighting artifacts like debris or small mitotic cells.

Main Results:

• CytoSummaryNet improved mechanism of action prediction by 30-68% in mean average precision compared to average profiling.
• Interpretability analysis revealed the model prioritizes large, uncrowded cells, effectively handling cellular heterogeneity.

Conclusions:

• CytoSummaryNet provides a straightforward post-processing method for single-cell profiles, significantly boosting retrieval performance.
• The approach requires only readily available perturbation labels for training, making it broadly applicable to image-based profiling datasets.