RJMCMC Nonlinear Survival Models

Julia implementation of reversible jump MCMC survival models for simulations and real-world analyses (German Breast Cancer and Primary Biliary Cirrhosis).

Features

• Three RJMCMC Methods:
  • NonLinear1: Default uniform order statistics prior
  • NonLinear2: Dirichlet-Gamma prior for knot locations
  • CoxPH: Baseline Cox proportional hazards model
• Parallel Execution: Thread-based parallelism with ordered execution (by g_type, then n_values)
• Resume Support: Automatic checkpointing and resume from existing results
• Progress Tracking: Real-time progress bar for long-running simulations
• Comprehensive Outputs: Posterior summaries, IBS metrics, and publication-ready plots

Project Structure

• config.jl: Central defaults for simulations (demo/full) and real-data runs
• model.jl: RJMCMC algorithms and utilities shared by all scripts
• simulation.jl: End-to-end simulation pipeline with caching, resume support, progress bar, and thread-based parallelism
• real_data_GBC.jl / real_data_PBC.jl: Analyses for the GBC and PBC datasets (CSV files in the project root)
• install_packages.jl: Script to install all required Julia packages
• table_gen.py: Generate LaTeX simulation tables from simu_summary.csv
• boxplot.R: R script for generating IBS boxplots (requires ggplot2)
• results/: Generated outputs (gitignored)

Installation

Prerequisites

• Julia >= 1.9: Download from julialang.org
• R (optional, for boxplot generation): Download from r-project.org

Install Julia Packages

Recommended method (easiest):

julia install_packages.jl

Alternative methods:

1. Windows PowerShell:

   julia -e "using Pkg; Pkg.add([\"DataFrames\", \"CSV\", \"Distributions\", \"ProgressMeter\", \"StatsBase\", \"VectorizedStatistics\", \"MLDataUtils\", \"Plots\", \"StatsPlots\", \"CategoricalArrays\", \"SpecialFunctions\", \"LaTeXStrings\"])"

2. Interactive Julia REPL:

   julia
   ] add DataFrames CSV Distributions ProgressMeter StatsBase VectorizedStatistics MLDataUtils Plots StatsPlots CategoricalArrays SpecialFunctions LaTeXStrings

(Press Ctrl+C to exit package mode, then type exit() to quit)

Install R Package (for boxplots)

Rscript -e "install.packages('ggplot2', repos='https://cran.rstudio.com/')"

Or in R console:

install.packages("ggplot2")

Running Simulations

Basic Usage

# Quick demo run (all scenarios: n=200,400,800, 5 replications each)
julia simulation.jl --demo

# Full run (all scenarios: n=200,400,800, 1000 replications each)
julia simulation.jl --full

# Use specific number of threads
JULIA_NUM_THREADS=32 julia simulation.jl --full

# Windows PowerShell
$env:JULIA_NUM_THREADS=32; julia simulation.jl --full

Command-Line Options

• --demo: Run demo mode (all scenarios with 5 replications each)
• --full: Run full mode (all scenarios with 1000 replications each)
• --reset: Clear existing checkpoints and rerun from scratch
• --replot: Regenerate plots from existing results
• --plot-only: Only generate plots without running simulations
• --workers=N: Specify number of worker threads (default: all available threads)

Execution Order

The simulation executes tasks in a specific order:

1. By g_type: linear -> quad -> sin
2. By n_values: Within each g_type, tasks are processed by sample size (200 -> 400 -> 800)
3. By replication: Within each (g_type, n) combination, replications are executed in parallel but tasks are sorted by replication index (1, 2, ..., N)

This ensures that all linear scenarios complete before starting quad, and all quad complete before starting sin.

Simulation Configuration

Demo Mode (--demo):

• Sample sizes: [200, 400, 800]
• Replications: 5 per scenario
• Total tasks: 3 g_types x 3 n_values x 5 replications = 45 tasks

Full Mode (--full):

• Sample sizes: [200, 400, 800]
• Replications: 1000 per scenario
• Total tasks: 3 g_types x 3 n_values x 1000 replications = 9000 tasks

Checkpoints and Resume

• Checkpoints are saved under results/simulation/<mode>/g=<type>/n=<size>/rep=<id>/results_dict.jls
• Rerunning will automatically skip finished tasks and continue from existing checkpoints
• Use --reset to clear all checkpoints and start fresh

Output Files

Results are saved in results/simulation/<mode>/:

• simu_summary.csv: Summary statistics for all methods
• df_IBS.csv: Integrated Brier Score (IBS) metrics
• plots/: Individual plots for each (g_type, n) combination
• plots_manuscript/: Publication-ready figures (Figures 1-4 style)

Real Data Analysis

# German Breast Cancer dataset
julia real_data_GBC.jl

# Primary Biliary Cirrhosis dataset
julia real_data_PBC.jl

Outputs are written to results/real_data/gbc/ and results/real_data/pbc/:

• beta_*.csv: Posterior summaries for NonLinear1, NonLinear2, and CoxPH (columns: Method, Beta, Pos_Mean, CrI_LB, CrI_UB)
• g_*.csv: Posterior mean of g(z) for NonLinear1 and NonLinear2 (columns: z, g_non1, g_non2)
• tau_*.csv, zeta_*.csv, HK_*.csv: Trace storage for tau, zeta, H, and K (NonLinear1 and NonLinear2)
• IBS_*.csv: Cross-validated IBS comparisons across methods
• g_compare_*.pdf: NonLinear1 vs NonLinear2 g(z) plot with labeled axes
• z_hist_*.pdf: Histogram of Z values used in the nonlinear term

Dependencies

See requirements.txt for a complete list of required packages:

• Julia packages: DataFrames, CSV, Distributions, ProgressMeter, StatsBase, VectorizedStatistics, MLDataUtils, Plots, StatsPlots, CategoricalArrays, SpecialFunctions, LaTeXStrings
• R packages: ggplot2 (for boxplot generation)
• Python packages: pandas (for table_gen.py)

Notes

• The simulation uses thread-based parallelism. Set JULIA_NUM_THREADS environment variable to control the number of threads.
• On Windows PowerShell, use $env:JULIA_NUM_THREADS=N to set thread count.
• The code automatically detects available threads and warns if requested threads exceed available threads.
• For best performance, set JULIA_NUM_THREADS to match your CPU core count.

Contact

If you have questions or run into issues, please contact zhanght@gdou.edu.cn.