Borderline CNVs in PGT-A: a simulation study

Overview

This repository contains reproducible simulation code for the paper:

Title: Borderline copy number variations in preimplantation genetic testing: when signal becomes noise Authors: Anere Cye Journal: Royal Society Open Science (2026) DOI: 10.5281/zenodo.19421746

Repository structure

• R/ - simulation functions
• scripts/ - run_all.R and scenario scripts
• results/ - generated automatically

How to run

Open R in the project root and run: source(scripts/run_all.R)

All simulations use set.seed(2025). 5000 replicates per condition.

WGS-PGT validation

The first-pass real-data workflow uses only one workbook and only two sheets:

• data/wgspgt/Fig.1d_Fig.1e_Fig.1f_Fig.1g_Fig.2b_Fig5.xlsx
• Fig1f_WGS for chromosome-level input
• Fig1d_Fig1e for embryo-level metadata

The loader in R/load_wgspgt_data.R:

• treats EmbryoNumber as embryo_id
• reshapes the WGS sheet from wide chromosome columns to tidy rows
• joins meanDepth, meanBreadth, EmbryoStatus, and Method
• skips all other sheets for now

Run the validation analysis from the project root with: Rscript analysis_wgspgt_validation.R

This first pass does not require a diagnosis sheet yet. It saves a cleaned dataset and descriptive threshold-based outputs to results/wgspgt_validation/:

• wgspgt_cleaned.csv
• wgspgt_threshold_calls.csv
• wgspgt_chromosome_summary.csv
• wgspgt_embryo_summary.csv
• wgspgt_method_summary.csv

Realistic simulation

The original simulation code is preserved as the idealized model. A new parallel prototype framework models a more realistic aggregated-signal path:

1. latent embryo state
2. latent chromosome-level signal
3. measurement / distortion / aggregation layer
4. embryo-level decision rule

Core files:

• MODEL_REDESIGN.md
• R/simulate_latent_embryo_state.R
• R/simulate_latent_chromosome_signal.R
• R/simulate_measurement_layer.R
• R/classify_embryo_realistic.R
• analysis_realistic_simulation.R

Run the realistic prototype from the project root with: Rscript analysis_realistic_simulation.R

Outputs are written to results/realistic_simulation/, including:

• chromosome-level observed signal tables
• embryo-level summaries
• calibration cutoffs for out-of-sample evaluation
• evaluation metrics for out-of-sample evaluation
• confusion matrices
• sensitivity / specificity / accuracy tables
• decision-rule comparison tables
• idealized vs realistic comparison tables
• signal distribution plots

Key results (Scenario B, n=5000)

False normal rate M=15%: 34.5% [33.2-35.9%] False positive rate diploid: 54.4% [53.0-55.8%]