Predicting cell lineages using autoencoders and optimal transport.

Area of Science:

• Computational Biology
• Cell Biology
• Bioinformatics

Background:

• Lineage tracing is crucial for understanding development and disease.
• Controlled time-course experiments are often not feasible for lineage tracing, especially with patient-derived samples.
• Existing methods struggle to reconstruct cell lineages without direct temporal observation.

Purpose of the Study:

• To develop a computational pipeline, ImageAEOT, for predicting cell lineages from static, time-labeled image datasets.
• To enable lineage tracing in scenarios where experimental time-course studies are impractical.
• To identify image-based features and biomarkers associated with cellular processes.

Main Methods:

• ImageAEOT utilizes autoencoders and optimal transport to predict cell lineages.
• The pipeline generates artificial lineages for cells based on population characteristics across different time stages.
• It analyzes single-cell images to infer relationships between cell populations.

Main Results:

• ImageAEOT was successfully applied to benchmark tasks involving fibroblast activation in 3D tissues.
• The method was validated on chromatin images from breast cancer cell lines and human tissue samples.
• It linked chromatin condensation patterns to tumor progression stages, demonstrating its utility in cancer research.

Conclusions:

• ImageAEOT offers a promising computational approach for cell lineage tracing in challenging experimental settings.
• The integration of autoencoders and optimal transport provides a novel solution for reconstructing cell histories.
• This method facilitates the discovery of image-based biomarkers for biological processes and diseases.