Skip to contents

Combine per-run genpca fits via meta-PCA (MFA)

fit_subject_metapca.Rd

Aggregates multiple genpca fits (e.g., per-run, per-session, or per-subject) into a single consensus component space using subpca::metapca with Multiple Factor Analysis (MFA) weighting. This is a “PCA of PCAs” that operates on the loadings from each individual fit, balancing contributions across inputs.

Usage

fit_subject_metapca(fits, k = 20)

Arguments

fits

List of genpca fit objects to combine.

k

Number of meta-components to return (consensus dimensionality).

Value

A metapca object with combined loadings v and related metadata.

Details

Two common workflows are supported in fmrigpca:

  1. Joint fit across runs: stack runs in time, build a block-diagonal row metric M, and fit once via fit_subject_genpca(). This shares a single spatial column metric A and exploits all timepoints jointly.

  2. Meta-PCA across individual fits: fit each run (or session/subject) separately to obtain genpca objects, then combine them with meta-PCA.

Meta-PCA uses MFA weighting to avoid any single fit dominating due to scale or sample size differences, and returns a metapca object with combined loadings (accessible as $v). Use this when:

  • Runs/sessions are heterogeneous and you prefer to fit them independently (e.g., different preprocessing, durations, or acquisition settings)

  • Combining across subjects after per-subject genpca fits

Assumptions and tips:

  • Loadings should be defined on the same voxel/parcel index set (e.g., same mask and order). If they differ, align to a common subset before fitting.

  • Keep k small and comparable across inputs; metapca selects a consensus space up to ncomp = k.

  • See the subpca package for algorithmic details of MFA combination.

See also

fit_subject_genpca() and fit_subject_genpca_parcel() for generating individual fits to combine

Other main fitting: fit_subject_genpca(), fit_subject_genpca_parcel()

Examples

# \donttest{
# Per-run fits (each list contains a single run)
fits <- lapply(seq_along(nv_list), function(r) {
  fit_subject_genpca(
    nv_list = list(nv_list[[r]]),
    mask_vol = mask,
    gm_vol = gm, wm_vol = wm, csf_vol = csf,
    k = 5, p_ar = 1L
  )
})
#> Error: object 'nv_list' not found
# Combine via meta-PCA (MFA weighting)
meta <- fit_subject_metapca(fits, k = 3)
#> Error: object 'fits' not found
str(meta$v)  # consensus loadings (voxels x k)
#> Error: object 'meta' not found
# }