Leveraging Lineage Barcodes as Natural Augmentations for Contrastive Learning of Cell Fate in scRNA-seq Data

Area of Science:

• Cellular biology
• Computational biology
• Genomics

Background:

• Understanding cell fate commitment is crucial for regenerative medicine and immunotherapy.
• Single-cell lineage tracing (scLT) is hindered by confounding factors like cell cycle, obscuring differentiation signals.

Purpose of the Study:

• To introduce Lineage-aware Contrastive Learning (LCL), a novel framework for isolating lineage-specific signals in scLT.
• To enable prediction of future cell-type compositions from early-stage data and model longitudinal fate commitment from cross-sectional datasets.

Main Methods:

• Developed LCL, a semi-supervised contrastive learning framework using inheritable lineage barcodes as natural data augmentation.
• Applied LCL to align unlabeled cells and transfer lineage structures to datasets lacking explicit barcoding.

Main Results:

• LCL effectively isolates subtle, lineage-specific signals by treating barcodes as data augmentation.
• Successfully predicted future cell-type compositions and modeled longitudinal fate commitment from cross-sectional data.
• Outperformed standard models like scVI in benchmarking studies on hematopoietic and fibroblast systems.

Conclusions:

• LCL provides a scalable and effective approach for deciphering cellular differentiation and fate commitment.
• Establishes contrastive learning as a powerful paradigm for understanding and potentially manipulating cellular differentiation processes.
• Facilitates the transfer of lineage information to clinical datasets, advancing precision therapeutics.