Skip to contents

Parallel Analysis on LDSC Results

Source: R/paLDSC.R
paLDSC.Rd

Determines the number of non-spurious latent factors via Monte Carlo eigenvalue comparison against simulated null distributions. Drop-in replacement for GenomicSEM::paLDSC(): prints the same "Parallel Analysis suggests extracting N components" banner and writes the same canonical PDF filenames (PA_LDSC.pdf, Diagonalized_PA_LDSC.pdf, FA_PA_LDSC.pdf, FA_Diagonalized_PA_LDSC.pdf) when save.pdf = TRUE. The return value is returned invisibly so interactive callers don't see the rich list dumped to the console, but if you assign it you get the structured fields described below.

Usage

paLDSC(
  S = S,
  V = V,
  r = NULL,
  p = NULL,
  save.pdf = FALSE,
  diag = FALSE,
  fa = FALSE,
  fm = NULL,
  nfactors = NULL,
  parallel = TRUE,
  cores = NULL
)

Arguments

S

Genetic covariance matrix (from LDSC output)

V

Sampling covariance matrix (from LDSC output)

r

Number of Monte Carlo simulations (default NULL = 500)

p

Percentile threshold for null distribution (default NULL = 0.95)

save.pdf

When TRUE, write the canonical PDF filenames matching R GenomicSEM::paLDSC to the working directory.

diag

Additionally run the analysis against diag(V) only and print/save the diagonalized variant (default FALSE).

fa

Additionally run psych factor analysis on the correlation matrix (default FALSE; requires psych).

fm

Factor method for psych::fa (e.g. "minres", "ml", "pa"; default NULL = "minres")

nfactors

Number of factors to extract when using fa (default NULL = auto from eigenvalue analysis)

parallel

Use a parallel rayon worker pool for the Monte Carlo simulation loop (default TRUE). Set to FALSE to force single-threaded execution.

cores

Integer cap on the rayon pool size. When NULL (the default) rayon honours RAYON_NUM_THREADS if set, else it uses the number of logical cores reported by the OS. On many-core machines (32+) or when the underlying BLAS is multithreaded, set this explicitly to avoid oversubscribing CPUs with nested BLAS threads.

Value

Invisibly, a list with components:

observed

Observed eigenvalues (descending)

simulated_95

Simulated eigenvalues at the given percentile

n_factors

Suggested number of factors from the eigenvalue-based parallel analysis

diag

(when diag=TRUE) the same three fields for the diagonalized variant

fa

(when fa=TRUE) loadings + uniquenesses from psych

Details

When fa=TRUE or fm is specified, uses psych package's factor analysis methods for the additional FA panels.

Examples

if (FALSE) { # \dontrun{
# `S` and `V` come from `ldsc()`:
pa <- paLDSC(S = covstruc$S, V = covstruc$V, r = 500, save.pdf = FALSE)
pa$n_factors   # number of non-spurious factors
pa$observed    # observed eigenvalues
} # }