Clonal embeddings allow exploratory analysis of lineage-resolved single-cell data.

Area of Science:

• Computational Biology
• Developmental Biology
• Immunology

Background:

• High-throughput lineage tracing coupled with single-cell transcriptomics offers deep insights into biological processes.
• Analyzing large-scale lineage-coupled single-cell RNA sequencing (scRNA-seq) data is challenging due to sparse clonal information and reliance on cell-type labels.

Purpose of the Study:

• To develop a machine learning approach, clone2vec, for robust analysis of lineage-coupled scRNA-seq data.
• To create informative clone embeddings that bypass discrete cell-type labels and handle sparse data.

Main Methods:

• Developed clone2vec, a machine learning method that learns clone embeddings from the cellular expression manifold.
• Applied clone2vec to prospective barcoding datasets from embryogenesis, tumorigenesis, and hematopoiesis.
• Utilized clone2vec for analyzing T-cell receptor (TCR) sequencing data from tumor microenvironments.

Main Results:

• clone2vec successfully recapitulates known clonal patterns and identifies novel continuous variation axes.
• The method implicates specific regulatory programs and developmental pathways in observed clonal variations.
• In tumor microenvironments, clone2vec accurately identifies distinct Treg lineages and conserved CD8+ T cell sublineages across cancer types.

Conclusions:

• clone2vec provides a stable and interpretable representation of clonal variation, summarizing data geometry effectively.
• This approach facilitates exploration, statistical analysis of clone-gene associations, and cross-dataset alignment.
• clone2vec offers a general and robust solution for the exploratory analysis of lineage-coupled scRNA-seq data.