Public API Reference

Layer

A struct representing a layer in the Neural Network Mixed Model (NNMM).

Fields

• layer_name::String: Name identifier for the layer
• data_path: Path to the data file(s) - String for single file, Vector{String} for partial connection
• separator: Column separator in the data file (default: ',')
• header: Whether the data file has a header row (default: true)
• data: Loaded data (initialized as empty)
• quality_control: Whether to perform quality control on genotypes (default: true)
• MAF: Minor Allele Frequency threshold for QC (default: 0.01)
• missing_value: Value representing missing data (default: 9.0)
• center: Whether to center the data (default: true)

Equation

A struct representing an equation connecting layers in NNMM.

Fields

• from_layer_name: Source layer name
• to_layer_name: Target layer name
• equation: Model equation string (e.g., "omics = intercept + geno")
• traits: Output trait column names (e.g., ["omic1", "omic2"] or ["trait1"])
• covariate: Covariate variable names
• random: Random effect specifications
• activation_function: Activation function - string ("linear", "sigmoid", "tanh", "relu", "leakyrelu") or custom Function
• partial_connect_structure: Structure for partial connectivity
• starting_value: Starting values for MCMC
• method: Bayesian method ("BayesC", "BayesA", "BayesB", etc.)
• Pi: Prior probability of zero effect
• estimatePi: Whether to estimate Pi
• Variance parameters for G (genetic) and R (residual)

Note: omics_name and phenotype_name are deprecated aliases for traits.

Omics

A struct representing omics data (e.g., transcriptomics, metabolomics) in NNMM.

Fields

• name: Name identifier for this omics category
• trait_names: Names for corresponding traits
• obsID: Row IDs for individuals
• featureID: Feature/omics variable IDs
• nObs: Number of observations
• nFeatures: Number of omics features
• nMarkers: Alias for nFeatures (compatibility)
• centered: Whether data is centered
• data: The omics data matrix
• ntraits: Number of traits in the model
• Additional fields for Bayesian MCMC computations

Phenotypes

A struct representing phenotype data in NNMM.

Fields

• obsID: Individual IDs
• featureID: Phenotype/trait names
• nObs: Number of observations
• nPheno: Number of phenotypes
• data: Phenotype data matrix
• nFeatures: Alias for nPheno

runNNMM(layers, equations; kwargs...)

Run Neural Network Mixed Model (NNMM) analysis.

Arguments

• layers: Vector of 3 Layer objects defining the network architecture
  • Layer 1: Genotypes (SNPs)
  • Layer 2: Omics/Latent traits
  • Layer 3: Phenotypes
• equations: Vector of 2 Equation objects defining relationships
  • Equation 1: Genotypes → Omics (1→2)
  • Equation 2: Omics → Phenotypes (2→3)

Keyword Arguments

MCMC Settings

• chain_length::Integer=100: Total MCMC iterations
• burnin::Integer=0: Number of burn-in iterations to discard
• output_samples_frequency::Integer: Save every nth sample (default: auto)
• output_prediction_frequency::Integer: Save every nth prediction sample for EBV/EPV (default: output_samples_frequency)
• update_priors_frequency::Integer=0: Update prior parameters every n iterations

Output Settings

• outputEBV=true: Output estimated breeding values
• output_heritability=true: Calculate heritability estimates
• output_folder="nnmm_results": Directory for output files

Computational Settings

• seed=false: Random seed for reproducibility
• double_precision=false: Use Float64 instead of Float32
• big_memory=false: Enable memory-intensive optimizations

Returns

A dictionary containing:

• Posterior means for all parameters
• MCMC samples (saved to files)
• EBV estimates

Example

# Define layers
layer1 = Layer(name="geno", file="genotypes.csv")
layer2 = Layer(name="omics", file="omics.csv")
layer3 = Layer(name="pheno", file="phenotypes.csv")

# Define equations
eq1 = Equation("omics = intercept + geno", 
               method="BayesC", omics_name=["o1","o2"])
eq2 = Equation("pheno = intercept + omics", 
               activation_function=tanh, phenotype_name=["y"])

# Run analysis
results = runNNMM([layer1, layer2, layer3], [eq1, eq2],
                  chain_length=5000, burnin=1000)

See also: Layer, Equation, describe

describe(model::MME)

Print a summary of the mixed model equations (MME) structure.

Displays:

• Model equations (truncated if >5)
• Term information (classification, fixed/random, number of levels)
• Prior distributions and hyperparameters
• Variance component settings
• Marker effect settings (if genomic data included)

Arguments

• model::MME: The mixed model equations object (created internally by runNNMM)

Output

Prints formatted model summary to stdout including:

• Model equations
• Term classification (covariate/factor, fixed/random)
• Prior settings for variance components
• MCMC configuration

Example

# After running runNNMM, the describe function is called automatically.
# For manual inspection:
describe(results["mme"])

read_phenotypes(file; separator=',', header=true, missing_value="NA", output_folder=".")

Read phenotype data from a CSV file.

Arguments

• file: Path to CSV file containing phenotype data

Keyword Arguments

• separator: Column delimiter (default: ',')
• header: Whether file has header row (default: true)
• missing_value: String representing missing values (default: "NA")
• output_folder: Directory for output files (default: current directory)

Returns

DataFrame with phenotype data. First column is individual IDs.

Example

pheno_df = read_phenotypes("phenotypes.csv", missing_value="NA")

Notes

• First column must contain individual IDs
• Missing values are converted to Julia's missing type
• A file IDs_for_individuals_with_phenotypes.txt is written to output_folder

nnmm_get_genotypes(file, G=false; kwargs...)

Read genotype data for the NNMM model (layer 1).

Arguments

• file: Genotype data source - can be:
  • String: Path to CSV file
  • DataFrame: DataFrame with ID column + marker columns
  • Matrix: Numeric matrix of genotypes
• G: Prior genetic variance (default: estimated from data)

Keyword Arguments

• method: Bayesian method ("BayesA", "BayesB", "BayesC", "RR-BLUP", "GBLUP", "BayesL")
• Pi: Prior inclusion probability (default: 0.0)
• estimatePi: Whether to estimate Pi (default: true)
• G_is_marker_variance: If true, G is marker variance; if false, G is genetic variance
• df: Degrees of freedom for prior (default: 4.0)
• quality_control: Perform QC filtering (default: true)
• MAF: Minor allele frequency threshold (default: 0.01)
• center: Center genotypes (default: true)
• separator: File delimiter (default: ',')
• header: File has header row (default: true)

Returns

Genotypes struct containing processed genotype data and method parameters.

get_genotypes(args...; kwargs...)

Alias for nnmm_get_genotypes. See that function for full documentation.

nnmm_get_omics(file, G=false; kwargs...)

Read omics data for the NNMM model (layer 2 / middle layer).

Arguments

• file: Path to CSV file containing omics data
• G: Prior genetic variance (default: estimated from data)

Keyword Arguments

• omics_name: Vector of column names to use as omics features (required)
• method: Bayesian method ("BayesA", "BayesB", "BayesC", "RR-BLUP", "BayesL")
• Pi: Prior inclusion probability (default: 0.0)
• estimatePi: Whether to estimate Pi (default: true)
• G_is_marker_variance: If true, G is marker variance; if false, G is genetic variance
• df: Degrees of freedom for prior (default: 4.0)
• constraint: Use independent variances for multi-trait (default: true)
• separator: File delimiter (default: ',')
• header: File has header row (default: true)
• missing_value: String/value representing missing data (default: false)

Returns

Omics struct containing the omics data and method parameters.

Example

omics = nnmm_get_omics("omics_data.csv", 
                       omics_name=["gene1", "gene2", "gene3"],
                       missing_value="NA")

Notes

• First column must be individual IDs
• Missing values will be sampled during MCMC (HMC for latent traits)
• Unlike genotypes, no quality control (MAF filtering) is applied

GWAS(marker_effects_file; header=true)

Compute the model frequency for each marker.

Model frequency is the probability that a marker is included in the model (i.e., has non-zero effect) across MCMC samples.

Arguments

• marker_effects_file: Path to CSV file with MCMC samples of marker effects
• header: Whether file has header row (default: true)

Returns

DataFrame with columns:

• marker_ID: Marker identifier
• modelfrequency: Proportion of samples where effect ≠ 0

Example

# After running NNMM with BayesB or BayesC
freq = GWAS("nnmm_results/MCMC_samples_marker_effects_genotypes.txt")

GWAS(model,map_file,marker_effects_file...;
     window_size = "1 Mb",sliding_window = false,
     GWAS = true, threshold = 0.001,
     genetic_correlation = false,
     header = true)

run genomic window-based GWAS

• MCMC samples of marker effects are stored in markereffectsfile with delimiter ','.
• model is either the model::MME used in analysis or the genotype cavariate matrix M::Array
• map_file has the (sorted) marker position information with delimiter ','. If the map file is not provided, i.e., map_file=false, a fake map file will be generated with window_size markers in each 1 Mb window, and each 1 Mb window will be tested.
• If two markereffectsfile are provided, and genetic_correlation = true, genomic correlation for each window is calculated.
• Statistics are computed for nonoverlapping windows of size window_size by default. If sliding_window = true, those for overlapping sliding windows are calculated.
• map file format:

markerID,chromosome,position
m1,1,16977
m2,1,434311
m3,1,1025513
m4,2,70350
m5,2,101135

getEBV(model::MME,traiti)

(internal function) Get breeding values for individuals defined by outputEBV(), defaulting to all genotyped individuals. This function is used inside MCMC functions for one MCMC samples from posterior distributions. e.g., non-NNBayespartial (multi-classs Bayes) : y1=M1α1[1]+M2α2[1]+M3α3[1] y2=M1α1[2]+M2α2[2]+M3α3[2]; NNBayespartial: y1=M1α1[1] y2=M2α2[1] y3=M3*α3[1];

dataset(file_name::AbstractString; dataset_name::AbstractString="")

Get the path to a built-in dataset file.

Arguments

• file_name::AbstractString: The name of the file to retrieve
• dataset_name::AbstractString="": Optional subdirectory name within the data folder

Returns

• String: Full path to the requested data file

Examples

phenofile = dataset("phenotypes.csv")
genofile = dataset("genotypes.txt", dataset_name="example")

get_pedigree(pedfile::AbstractString;header=false,separator=',',missingstrings=["0"],output_folder=".")

• Get pedigree informtion from a pedigree file with header (defaulting to false) , separator (defaulting to ,) and missing values (defaulting to ["0"])
• output_folder: Directory to save diagnostic files (defaulting to current directory)
• Pedigree file format:

a,0,0
c,a,b
d,a,c