Muscle Synergy Decomposition

Decomposes multi-channel EMG into muscle synergies using matrix factorization.

Usage

muscleSynergy(
  x,
  n_synergies,
  method = c("nmf", "pca", "ica"),
  n_restarts = 1L,
  max_iter = 200L,
  tol = 1e-04,
  seed = NULL,
  assay_name = NULL
)

Arguments

x

A PhysioExperiment object with multi-channel EMG.

n_synergies

Number of synergies to extract.

method

Decomposition method: "nmf" (non-negative matrix factorization), "pca" (principal component analysis), or "ica" (independent component analysis).

n_restarts

Number of random NMF restarts; the fit with the lowest reconstruction error (highest VAF) is returned (default: 1). Ignored for "pca".

max_iter

Maximum iterations for NMF (default: 200).

tol

Convergence tolerance for NMF (default: 1e-4).

seed

Optional integer seed for reproducible random initialization (NMF restarts and ICA). If set, the returned best-of-restarts solution is deterministic.

assay_name

Input assay name (default: first assay).

Value

A list with:

• W: Synergy weight matrix (n_synergies x channels)

• H: Activation pattern matrix (time x n_synergies)

• vaf: Variance accounted for (0-1)

• method: Method used

• n_restarts: Number of restarts used

• all_vaf: VAF of each NMF restart (NA for pca/ica)

• convergence: For NMF, a list with the best restart's iterations and converged flag (NULL otherwise)

• original_data: Original data matrix for reconstruction

References

Lee, D.D. & Seung, H.S. (1999). "Learning the parts of objects by non-negative matrix factorization." Nature, 401(6755), 788-791. doi:10.1038/44565

Tresch, M.C., Cheung, V.C.K. & d'Avella, A. (2006). "Matrix factorization algorithms for the identification of muscle synergies." Journal of Neurophysiology, 95(4), 2199-2212. doi:10.1152/jn.00222.2005

See also

muscleSynergyOrder() for model-order (synergy count) selection, synergyReconstruct() for reconstructing data from synergies, synergyCompare() for comparing synergy solutions