MARGARET

MARGARET reconstructs single-cell RNA-seq trajectory topologies and quantifies cell-fate plasticity to elucidate cellular developmental processes such as differentiation and fate mapping.

Key Features:

• Deep unsupervised metric learning: Learns cell–cell similarity metrics from scRNA-seq data using a deep unsupervised metric learning approach.
• Graph-partitioning with connectivity measure: Partitions the learned cell graph using a novel connectivity measure to recover complex topology.
• Trajectory topology reconstruction: Reconstructs intricate trajectory topologies including branching and multifurcations.
• Terminal state detection: Automatically detects terminal cell states from the inferred topology.
• Fate plasticity quantification: Generalizes quantification of cell-fate plasticity across complex cellular processes.
• Pseudotime ordering: Orders cells along pseudotime to recover global topology and temporal progression.
• Performance and benchmarking: Demonstrates robust performance on synthetic and real datasets, outperforming state-of-the-art methods in topology recovery and cell ordering.
• Scalability: Scales to large scRNA-seq datasets comprising millions of cells.

Scientific Applications:

• Human hematopoiesis: Precisely identified major lineages, correlated gene expression trends, and pinpointed transitional progenitors at branching points.
• Embryoid body differentiation: Revealed novel transitional populations that were subsequently validated by bulk sequencing.
• Mesoderm lineage characterization: Functionally distinguished different precursor populations within mesoderm lineages.
• Colon differentiation and disease: Characterized BEST4/OTOP2 cell lineages and detailed goblet cell heterogeneity under normal and inflamed conditions associated with ulcerative colitis.
• Benchmarking on synthetic and real datasets: Applied to synthetic and real scRNA-seq datasets to evaluate topology recovery and pseudotime ordering.

Methodology:

Implements deep unsupervised metric learning combined with a graph-partitioning strategy based on a novel connectivity measure, with automatic terminal state detection, pseudotime ordering, and quantification of fate plasticity.