CAPE: Covariance Alignment for Population Embedding in Single-cell Studies

Overview

CAPE is a sample-level embedding framework for single-cell cohort studies. It compares whole samples by aligning the covariance structures of their gene expression matrices. It does not require cell-type annotations or precomputed cell embeddings.

Key highlights

• Developed the Sample Conformity Metric to quantify sample similarity through spectral covariance alignment.
• Built a spectral embedding to reveal population structures, including disease trajectories.
• Identified genes whose expression distributions shift along the inferred trajectories.
• Evaluated metric robustness under batch-effect perturbations using statistical simulations.
• Packaged CAPE as a Scanpy-compatible Python library with support for resumable computation.