Trajectory-guided dimension reduction for multi-sample single-cell RNA-seq data

Usage

mustard(
  expr,
  pseudotime,
  cellanno = NULL,
  interval = NULL,
  r = 3,
  resolution = 101,
  smooth = 0.001,
  maxiter = 20,
  epsilon = 1e-04
)

Arguments

expr: The normalized and standardized gene expression matrix. Rows represent genes and columns represent cells
pseudotime: The vector of user-provided pseudotime values
cellanno: The vector indicating which sample each cell belongs to
interval: Range of pseudotime (range of the original pseudotime by default)
r: Number of components to decompose into (3 by default)
resolution: Number of pseudotime values to evaluate in the temporal loading function (101 by default). It does not affect the sample or gene loading vector
smooth: Smoothing parameter for RKHS norm (1e-3 by default). Larger value means smoother temporal loading functions
maxiter: Maximum number of iterations (20 by default)
epsilon: Convergence criteria for difference between iterations (1e-4 by default)

Value

A list including the estimated loadings and explained variances

A.hat: Sample loading matrix
B.hat: Gene loading matrix
Phi.hat: Temporal loading matrix
time: The pseudotime values where the temporal loading function is evaluated
Lambda: Eigenvalue vector
r.square: Variance explained by each component, which is the R-squared of the linear regression of the vectorized temporal tensor against the vectorized low-rank reconstruction using each component
accum.r.square: Variance explained by the top components accumulated, which is the R-squared of the linear regression of the vectorized temporal tensor against the vectorized low-rank reconstruction using the top components