JWAS is a Julia software platform for analyses of univariate and multivariate Bayesian mixed effects models. These models support routine single-trait and multi-trait genomic prediction and genome-wide association studies using complete, streaming, and incomplete genomic data workflows. JWAS provides Bayesian whole-genome methods including shrinkage estimation, variable selection, annotation-aware marker priors, and dense or block genotype workflows. The features of JWAS include:

• Univariate (single-trait) analysis
• Multivariate (multi-trait) analysis
• No limitations on fixed effects (e.g., herd, year, age, sex)
• Random effects other than markers (e.g., litter, pen)
• Random effects using pedigree information
  • Additive genetic effects
  • Maternal effects
• Random permanent environmental effects
• Correlated residuals
• Correlated random effects
• Unknown (or known) variance components
• Use of genomic information
  • Complete genomic data
  • Incomplete genomic data (single-step)
  • Dense genotype matrices
  • Streaming genotype storage for large marker panels
  • Exact fast-block sampling with fast_blocks
  • Approximate independent block sampling with independent_blocks=true
• Bayesian whole-genome marker models
  • BayesA, BayesB, BayesC, and BayesR workflows
  • Annotated BayesC and Annotated BayesR
  • Dense 2-trait Annotated BayesC

Supporting and Citing

We hope the friendly user interface and fast computing speed of JWAS will provide power and convenience for users in both industry and academia to analyze large datasets. Further, as a well-documented open-source software tool, we hope JWAS will also be used by a group of active community members, who will contribute to the source code and help maintain the project. Junior scientists can understand and learn the methodologies for whole-genome analyses by using JWAS and reading the tutorials and source code.

If you would like to help support JWAS, please star the repository on the upper right corner here as such statistic will help to demonstrate the active involvement of the community. If you use JWAS for your research, teaching, or other activities, we would be grateful if you could cite our work following Citing.

The trouble, the error and the new feature

If you have trouble using JWAS, want new features or find errors in JWAS, please post it in our discussion group, open an issue, or contact <qtlcheng@ucdavis.edu>.

Tutorials

Theory

• Some Theory in JWAS
• • A Table for Bayesian Linear Mixed Models (BLMM)
  • Models
  • • Complete Genomic Data
    • Incomplete Genomic Data
    • Priors

Manual

• Get Started
• • Installation
  • Access documentation
  • Run your analysis
• Workflow
• • Method-Specific Guides
  • Available Models
  • Get Data Ready
  • Step 1: Load Packages
  • Step 2: Read Data
  • Step 3: Build Model Equations
  • Step 4: Set Factors or Covariate
  • Step 5: Set Random or Fixed Effects
  • Step 6: Run Bayesian Analysis
  • check results
  • • Output files:
• BayesC and BayesR Comparison (Single-Trait)
• • Starting Values
  • Starting Value Sources
  • Variance-Scale Starting Values
  • Prior Structure
  • Sampler Order
  • Key Distinctions
  • Related Pages
  • Multi-Trait Note
• Annotated BayesC
• • Method Overview
  • Input Requirements
  • Dense Example
  • Multi-Trait Method
  • Dense Example with fast_blocks
  • Streaming Example
  • Practical Notes
• Multi-Trait Annotated BayesC
• • Support Boundaries
  • Method Overview
  • Input Requirements
  • Startup Behavior
  • Sampler Selection
  • Example
  • Output Interpretation
  • Practical Notes
• Annotated BayesR
• • Method Overview
  • Initialization
  • Sampler Order
  • Annotation Update In Detail
  • Latent Probit Form
  • Liability Update
  • Coefficient Update
  • Annotation-Effect Variance Update
  • Prior Reconstruction
  • Pseudocode
  • Input Requirements
  • Dense Example
  • Dense Block Example
  • Output
  • Practical Notes
• Block BayesC (fast_blocks)
• • When This Path Is Used
  • • Quick Semantics Checklist
  • Single-Trait BayesC Without Blocks
  • Single-Trait Block BayesC in JWAS
  • • Precomputation (once before MCMC)
    • Update flow inside each MCMC outer iteration
    • Important implementation detail
  • Exact Sequential Blocks vs Independent Blocks
  • • Single-Trait View
    • Multi-Trait View
    • Server Use
    • Detailed Comparison with BayesR3
  • Algorithm Comparison
  • Computational Complexity
  • • Standard BayesC (non-block)
    • JWAS block BayesC (current implementation)
    • BayesR3 (block strategy and paper fit)
    • Practical differences in complexity interpretation
    • Numerical example (N=200,000, P=2,000,000)
  • Detailed Resource Model (Current fast_blocks Path)
  • • Memory formulas
    • Runtime working set
    • I/O and precompute considerations
  • Worked Large-Scale Example (N=500,000, P=2,000,000)
  • What To Watch Closely
  • Example: Speed/Memory Tradeoff
  • Practical Guidance
• Benchmark
• • Goal
  • Environment
  • Fast-Block Benchmark (fast_blocks=true)
  • • Design
    • Measured t_mcmc (seconds)
    • Extrapolated Target Time
  • Standard BayesC Benchmark (fast_blocks=false)
  • • Design
    • Measured t_mcmc (seconds)
    • Extrapolated Target Time
  • Summary Comparison
  • Notes and Interpretation
• Prototype-to-Production Benchmarking Guidelines
• • When To Use This Page
  • Core Principles
  • • 1. Benchmark the production path first
    • 2. Separate deterministic correctness from stochastic agreement
    • 3. Hold the benchmark inputs constant before blaming the sampler
    • 4. Do not trust a single short MCMC run
    • 5. Save enough artifacts that someone else can rerun the benchmark
  • Benchmark Types Required For JWAS
  • • A. Smoke benchmark
    • B. Deterministic setup benchmark
    • C. One-step controlled replay benchmark
    • D. Production-vs-local benchmark
    • E. Long-chain parity benchmark
    • F. Multiseed parity benchmark
    • G. Performance and scale benchmarks
  • Recommended Workflow
  • • Step 1: Define the production target
    • Step 2: Build a shared benchmark dataset
    • Step 3: Run the production smoke benchmark
    • Step 4: Align deterministic setup
    • Step 5: Run a one-step controlled replay
    • Step 6: Compare production JWAS against local Julia diagnostics
    • Step 7: Run long-chain parity
    • Step 8: Run multiseed parity
    • Step 9: Write the benchmark report
  • Required Artifacts
  • Common Failure Modes
  • • Comparing raw stochastic traces across languages without shared draws
    • Concluding “bug” from one short chain
    • Benchmarking only a local helper instead of production runMCMC
    • Comparing only one metric
    • Changing preprocessing between implementations
    • Mixing correctness and performance questions
  • Acceptance Checklist
  • Worked Example: BayesR
  • Practical Recommendation
• Memory Usage (BayesA/B/C Marker Paths)
• • Notation
  • Non-Block Mode (fast_blocks=false)
  • • What is stored
    • Approximate memory formulas
  • Block Mode Additions (fast_blocks != false)
  • Worked Example (N=500,000, P=2,000,000)
  • • Original non-block version (fast_blocks=false)
    • Extra objects introduced by block mode (fast_blocks != false)
  • Practical Takeaways
• Handling Large Genotype Data Without Loading the Full Matrix into Memory
• • Scope
  • Current JWAS Status
  • • Streaming MVP workflow
    • Streaming MVP constraints
  • Notation
  • Baseline Complexity (Original BayesC)
  • • Rough operation count
  • Baseline Memory (Original BayesC)
  • Worked Example (N=500,000, P=2,000,000)
  • • Memory totals for original BayesC
  • Out-of-Core Working-Set Math (Original BayesC)
  • Representation Approaches for Original BayesC
  • • 1. Dense In-Memory (current baseline)
    • 2. Dense Out-of-Core (mmap-style dense backend)
    • 3. Native Bit-Packed Backend (recommended long-term)
    • 4. External Adapter Backend (e.g., PLINK-style reader)
  • Fast-Block Implementation (fast_blocks) in Large-Data Context
  • What To Watch for fast_blocks
  • Speed and I/O Considerations
  • • I/O lower-bound model
  • Side-by-Side Summary
  • Validation Details Needed for Large-Data Streaming
  • Practical Takeaways
• Streaming Genotype Backend (BayesC MVP)
• • Status and Design
  • Supported Scope (MVP)
  • Public API
  • • prepare_streaming_genotypes(...)
    • How to use conversion_mode
    • get_genotypes(...; storage=:stream)
  • End-to-End Example
  • Benchmark Snapshot
  • • 1) Correctness: simulated_omics dense vs stream
    • 2) Large synthetic benchmark (N=10,000, P=5,000)
    • 2b) Conversion-phase benchmark (N=10,000, P=5,000)
    • 3) Target-scale feasibility estimator (N=500,000, P=2,000,000)
  • Notes
• Streaming Genotype Workflow: A Conceptual Walkthrough
• • 1. The Problem
  • 2. Input: Text Genotype File
  • 3. One-Time Conversion: prepare_streaming_genotypes
  • 4. Loading: get_genotypes(prefix; storage=:stream)
  • 5. MCMC: Per Iteration, Per Marker
  • 6. Initial yCorr
  • 7. MVP Constraints
  • 8. End-to-End Flow
  • 9. Memory Comparison
• Public functions
• • Index
  • Public Interface
• Internal functions
• • Index
  • Internal interface

Examples

• Examples