mirror of
https://github.com/MillironX/beefblup.git
synced 2024-11-14 11:33:10 +00:00
287 lines
7.8 KiB
Julia
287 lines
7.8 KiB
Julia
# beefblup
|
|
# Main script for performing single-variate BLUP to find beef cattle
|
|
# breeding values
|
|
# Usage: julia beefblup.jl
|
|
# (C) 2020 Thomas A. Christensen II
|
|
# Licensed under BSD-3-Clause License
|
|
|
|
# Import the required packages
|
|
using XLSX
|
|
using LinearAlgebra
|
|
using Dates
|
|
using Gtk
|
|
|
|
# Display stuff
|
|
println("beefblup v 0.1")
|
|
println("(C) 2020 Thomas A. Christensen II")
|
|
println("https://github.com/millironx/beefblup")
|
|
print("\n")
|
|
|
|
### Prompt User
|
|
# Ask for an input spreadsheet
|
|
path = open_dialog_native(
|
|
"Select a beefblup worksheet",
|
|
GtkNullContainer(),
|
|
("*.xlsx", GtkFileFilter("*.xlsx", name="beefblup worksheet"))
|
|
)
|
|
|
|
# Ask for an output text filename
|
|
savepath = save_dialog_native(
|
|
"Save your beefblup results",
|
|
GtkNullContainer(),
|
|
(GtkFileFilter("*.txt", name="Results file"),
|
|
"*.txt")
|
|
)
|
|
|
|
# Ask for heritability
|
|
print("What is the heritability for this trait?> ")
|
|
h2 = parse(Float64, readline(stdin))
|
|
|
|
### Import input filename
|
|
print("[🐮]: Importing Excel file...")
|
|
|
|
# Import data from a suitable spreadsheet
|
|
data = XLSX.readxlsx(path)[1][:]
|
|
|
|
print("Done!\n")
|
|
|
|
### Process input file
|
|
print("[🐮]: Processing and formatting data...")
|
|
|
|
# Extract the headers into a separate array
|
|
headers = data[1,:]
|
|
data = data[2:end,:]
|
|
|
|
# Sort the array by date
|
|
data = sortslices(data, dims=1, lt=(x,y)->isless(x[2],y[2]))
|
|
|
|
# Define fields to hold id values for animals and their parents
|
|
ids = string.(data[:,1])
|
|
damids = string.(data[:,3])
|
|
sireids = string.(data[:,4])
|
|
numanimals = length(ids)
|
|
|
|
# Find the index values for animals and their parents
|
|
dam = indexin(damids, ids)
|
|
sire = indexin(sireids, ids)
|
|
|
|
# Store column numbers that need to be deleted
|
|
# Column 6 contains an intermediate Excel calculation and always need to
|
|
# be deleted
|
|
colstokeep = [1, 2, 3, 4, 5]
|
|
|
|
# Find any columns that need to be deleted
|
|
for i in 7:length(headers)
|
|
if length(unique(data[:,i])) <= 1
|
|
colname = headers[i]
|
|
print("Column '")
|
|
print(colname)
|
|
print("' does not have any unique animals and will be removed from this analysis\n")
|
|
else
|
|
push!(colstokeep, i)
|
|
end
|
|
end
|
|
|
|
# Delete the appropriate columns from the datasheet and the headers
|
|
data = data[:, colstokeep]
|
|
headers = headers[colstokeep]
|
|
|
|
# Determine how many contemporary groups there are
|
|
numgroups = ones(1, length(headers)-5)
|
|
for i in 6:length(headers)
|
|
numgroups[i-5] = length(unique(data[:,i]))
|
|
end
|
|
|
|
# If there are more groups than animals, then the analysis cannot continue
|
|
if sum(numgroups) >= numanimals
|
|
println("There are more contemporary groups than animals. The analysis will
|
|
now abort.")
|
|
exit()
|
|
end
|
|
|
|
# Define a "normal" animal as one of the last in the groups, provided that
|
|
# all traits do not have null values
|
|
normal = Array{String}(undef,1,length(headers)-5)
|
|
for i in 6:length(headers)
|
|
for j in numanimals:-1:1
|
|
if !ismissing(data[j,i])
|
|
normal[i-5] = string(data[j,i])
|
|
break
|
|
end
|
|
end
|
|
end
|
|
|
|
print("Done!\n")
|
|
|
|
### Create the fixed-effect matrix
|
|
print("[🐮]: Creating the fixed-effect matrix...")
|
|
|
|
# Form the fixed-effect matrix
|
|
X = zeros(Int8, numanimals, floor(Int,sum(numgroups))-length(numgroups)+1)
|
|
X[:,1] = ones(Int8, 1, numanimals)
|
|
|
|
# Create an external counter that will increment through both loops
|
|
counter = 2
|
|
|
|
# Store the traits in a string array
|
|
adjustedtraits =
|
|
Array{String}(undef,floor(Int,sum(numgroups))-length(numgroups))
|
|
# Iterate through each group
|
|
for i in 1:length(normal)
|
|
# Find the traits that are present in this trait
|
|
localdata = string.(data[:,i+5])
|
|
traits = unique(localdata)
|
|
# Remove the normal version from the analysis
|
|
effecttraits = traits[findall(x -> x != normal[i], traits)]
|
|
# Iterate inside of the group
|
|
for j in 1:length(effecttraits)
|
|
matchedindex = findall(x -> x != effecttraits[j], localdata)
|
|
X[matchedindex, counter] .= 1
|
|
# Add this trait to the string
|
|
adjustedtraits[counter - 1] = traits[j]
|
|
# Increment the big counter
|
|
global counter = counter + 1
|
|
end
|
|
end
|
|
|
|
print("Done!\n")
|
|
|
|
### Additive relationship matrix
|
|
print("[🐮]: Creating additive relationship matrix...")
|
|
|
|
# Create an empty matrix for the additive relationship matrix
|
|
A = zeros(numanimals, numanimals)
|
|
|
|
# Create the additive relationship matrix by the FORTRAN method presented by
|
|
# Henderson
|
|
for i in 1:numanimals
|
|
if !isnothing(dam[i]) && !isnothing(sire[i])
|
|
for j in 1:(i-1)
|
|
A[j,i] = 0.5*(A[j,sire[i]] + A[j,dam[i]])
|
|
A[i,j] = A[j,i]
|
|
end
|
|
A[i,i] = 1 + 0.5*A[sire[i], dam[i]]
|
|
elseif !isnothing(dam[i]) && isnothing(sire[i])
|
|
for j in 1:(i-1)
|
|
A[j,i] = 0.5*A[j,dam[i]]
|
|
A[i,j] = A[j,i]
|
|
end
|
|
A[i,i] = 1
|
|
elseif isnothing(dam[i]) && !isnothing(sire[i])
|
|
for j in 1:(i-1)
|
|
A[j,i] = 0.5*A[j,sire[i]]
|
|
A[i,j] = A[j,i]
|
|
end
|
|
A[i,i] = 1
|
|
else
|
|
for j in 1:(i-1)
|
|
A[j,i] = 0
|
|
A[i,j] = 0
|
|
end
|
|
A[i,i] = 1
|
|
end
|
|
end
|
|
|
|
print("Done!\n")
|
|
|
|
### Perform BLUP
|
|
print("[🐮]: Solving the mixed-model equations...")
|
|
|
|
# Extract the observed data
|
|
Y = convert(Array{Float64}, data[:,5])
|
|
|
|
# The random effects matrix
|
|
Z = Matrix{Int}(I, numanimals, numanimals)
|
|
|
|
# Remove items where there is no data
|
|
nullobs = findall(isnothing, Y)
|
|
Z[nullobs, nullobs] .= 0
|
|
|
|
# Calculate heritability
|
|
λ = (1-h2)/h2
|
|
|
|
# Use the mixed-model equations
|
|
MME = [X'*X X'*Z; Z'*X (Z'*Z)+(inv(A).*λ)]
|
|
MMY = [X'*Y; Z'*Y]
|
|
solutions = MME\MMY
|
|
|
|
# Find the accuracies
|
|
diaginv = diag(inv(MME))
|
|
reliability = ones(Float64, length(diaginv)) - diaginv.*λ
|
|
|
|
print("Done!\n")
|
|
|
|
### Output the results
|
|
print("[🐮]: Saving results...")
|
|
|
|
# Find how many traits we found BLUE for
|
|
numgroups = numgroups .- 1
|
|
|
|
# Start printing results to output
|
|
fileID = open(savepath, "w")
|
|
write(fileID, "beefblup Results Report\n")
|
|
write(fileID, "Produced using beefblup for Julia (")
|
|
write(fileID, "https://github.com/millironx/beefblup")
|
|
write(fileID, ")\n\n")
|
|
write(fileID, "Input:\t")
|
|
write(fileID, path)
|
|
write(fileID, "\nAnalysis performed:\t")
|
|
write(fileID, string(Dates.today()))
|
|
write(fileID, "\nTrait examined:\t")
|
|
write(fileID, headers[5])
|
|
write(fileID, "\n\n")
|
|
|
|
# Print base population stats
|
|
write(fileID, "Base Population:\n")
|
|
for i in 1:length(numgroups)
|
|
write(fileID, "\t")
|
|
write(fileID, headers[i+5])
|
|
write(fileID, ":\t")
|
|
write(fileID, normal[i])
|
|
write(fileID, "\n")
|
|
end
|
|
write(fileID, "\tMean ")
|
|
write(fileID, headers[5])
|
|
write(fileID, ":\t")
|
|
write(fileID, string(solutions[1]))
|
|
write(fileID, "\n\n")
|
|
|
|
# Contemporary group adjustments
|
|
counter = 2
|
|
write(fileID, "Contemporary Group Effects:\n")
|
|
for i in 1:length(numgroups)
|
|
write(fileID, "\t")
|
|
write(fileID, headers[i+5])
|
|
write(fileID, "\tEffect\tReliability\n")
|
|
for j in 1:numgroups[i]
|
|
write(fileID, "\t")
|
|
write(fileID, adjustedtraits[counter - 1])
|
|
write(fileID, "\t")
|
|
write(fileID, string(solutions[counter]))
|
|
write(fileID, "\t")
|
|
write(fileID, string(reliability[counter]))
|
|
write(fileID, "\n")
|
|
|
|
global counter = counter + 1
|
|
end
|
|
write(fileID, "\n")
|
|
end
|
|
write(fileID, "\n")
|
|
|
|
# Expected breeding values
|
|
write(fileID, "Expected Breeding Values:\n")
|
|
write(fileID, "\tID\tEBV\tReliability\n")
|
|
for i in 1:numanimals
|
|
write(fileID, "\t")
|
|
write(fileID, ids[i])
|
|
write(fileID, "\t")
|
|
write(fileID, string(solutions[i+counter-1]))
|
|
write(fileID, "\t")
|
|
write(fileID, string(reliability[i+counter-1]))
|
|
write(fileID, "\n")
|
|
end
|
|
|
|
write(fileID, "\n - END REPORT -")
|
|
close(fileID)
|
|
|
|
print("Done!\n")
|