PRS.jl Documentation

PlinkReader(path; markerIndex = false, sampleIndex = false)

Structure to access PLINK .bed, .bim, and .fam files at path.

Arguments

• markerIndex::Bool should index marker -> idx be created?
• sampleIndex::Bool should index sample -> idx be created?

Sample(fid, iid, father, mother, sex)

PLINK sample info

nsamples(p::PlinkReader)

Number of samples in Plink file.

samples(p::PlinkReader)

Return Array of Sample structs for samples in Plink file.

sample_index(p::PlinkReader, iid::String)

Return index of sample iid in PlinkReader. Uses index if available, otherwise linear search.

Marker(chrom, id, cm, pos, a1, a2)

PLINK marker info

nmarkers(p::PlinkReader)

Number of markers in Plink file.

markers(p::PlinkReader)

Return array of Marker structs for markers in Plink file.

marker_index(p::PlinkReader, id::String)

Return index of marker id in PlinkReader. Uses index if available, otherwise linear search.

markersDF(p::PlinkReader)

Get data frame with marker info.

Returns data frame with columns

• Chrom
• Name
• cM
• Pos
• A1
• A2
• Idx

getindex(p::PlinkReader, s::Int, m::Int)

Retrieve genotype of sample s and marker m.

Plink convention for representation of genotypes is as follows:

• 0b00 Hom1
• 0b01 Het
• 0b10 missing
• 0b11 Hom2

dosageMatrix(p::PlinkReader, markerIdx, sampleIdx = nothing;
               normalize = true)

Create dosage matrix (samples x markers) for markerIdx and sampleIdx from PlinkReader. Here dosage is expressed as count of alternative allele.

Replace missing genotypes with mean dosage. If normalize = true, normalize markers to mean μ = 0 and standard deviation σ = 1.

LDMatrix(p::PlinkReader, mIdx0, mIdx1; alpha = 0.9, window = posWindow(1_000_000))

Compute LDMatrix from genotypes in Plink file.

LDMatrix(path::AbstractString)

Load LDMatrix from .bim and .lds files at path.

markers(ld::LDMatrix)

Get array of Marker structs for all markers in LDMatrix ld.

markersDF(ld::LDMatrix)

Get data frame with marker info.

Returns data frame with columns

• Chrom
• Name
• cM
• Pos
• A1
• A2
• Idx

save(ld::LDMatrix, path::AbstractString)

Save LDMatrix to path.

ldscore(ld::LDMatrix)

Compute LD scores for markers in LDMatrix ld.

ldpred_gibbs(z, D, p, σ2, μ0; n_burnin = 100, n_iter = 500, verbose = false)

Run LDpred Gibbs sampler.

Arguments

• z::Vector: z-scores
• D::LDMatrix: LD matrix
• p::Real : proportion of variants deemed to be causal
• σ2::Real: $Nh²/M$ as estimated from LDscore regression (estimate_h2, or mean(z^2)/mean(lds)) Note: prior variance of non-null component of $μ$, $σ2_μ = σ2/p$
• μ0::Vector: starting estimate (e.g. from infinite model)

estimate_h2(z, lds, n)

Estimate total heritability $h^2$ for markers using LD score regression with intercept forced to 1.

Arguments

• lds::Vector: LD score as computed by ldscore(LDMatrix)
• n::Integer: effective number of samples ($4 n_0*n_1/(n_0+n_1)$ for case-control)

estimate_neff(beta, sebeta, freq)

Estimate effective number of samples in case-control study.

Effective number of samples is total number of samples for a cohort with same number of samples and controls, that would result in the observed sebeta.

freq can be minor or major allele freq, as only freq*(1-freq) is being used.

z2cor(z, n)

Convert GWAS z-value to phenotype-genotype correlation coefficient.

elnetg_path(X, r, λs, s;
            maxiter = 10_000, thresh = 1e-4)

Solve elastic net for path along λs using warm starts.

Arguments

• X::Matrix nsubj x nmarkers normalized genotype matrix column normalized (μ = 0, σ = 1)
• r::Vector nmarkers x 1 vector of correlation coefficients between phenotype and genotypes
• λs::Vector shrinkage parameter path for 1-norm in decreasing order
• s::Real shrinkage parameter for 2-norm (LD)
• β::Vector warm start (in) and result (out) for solution vector
• maxiter::Int maximum number of iterations
• thresh::Real maximum change in β to be called converged

snp_join(df1, df2; on = [:Chrom, :Pos],
         alleles1 = [:A1_1, :A2_1], alleles2 = [:A1_2, :A2_2],
         matchcol = :sign)

Join data frames df1 and df2 and match variants.

Data frames df1 and df2 are first matched by columns specified in on. Then column matchcol is set to

• matchcol == 1 if values for alleles1 and alleles2 match
• matchcol == -1 if values for alleles1 and alleles2 are swapped
• matchcol == 0 otherwise

Note: Swapped means simply swapping of alleles, not applying reverse complement as done in other implementations.