Potato_GS_pipeline

Genomic selection (GS) pipeline for potato breeding: within- and across-environment prediction, genotyping platform comparison, and breeding program simulation using AlphaSimR.

---

Repository Structure

Potato_GS_pipeline/
│
├── data/                            # to be added
│
├── codes/
│   ├── CV1_within_env/              # Section 3.2 — Within-environment prediction
│   │   └── GS_CV1_GBLUP.R
│   │
│   ├── CV0_across_env/              # Section 3.3 — Across-environment prediction
│   │   └── GS_CV0_LOEO.R
│   │
│   ├── platform_comparison/         # Section 3.4 — Genotyping platform comparison
│   │   └── GS_CV1_DArT_4K.R
│   │
│   ├── simulation/                  # Section 3.5 — AlphaSimR breeding simulation
│   │   ├── RUNME.R                  #   Main orchestrator
│   │   ├── GlobalParameters.R       #   Trait architecture & base population
│   │   ├── FillPipeline.R           #   Initialize breeding pipeline
│   │   ├── AdvanceBurnin.R          #   Burn-in advancement (phenotypic)
│   │   ├── Advance_Conventional.R   #   Scenario: Conventional selection
│   │   ├── Advance_Strategy1.R      #   Scenario: Genotype 650 seedlings
│   │   ├── Advance_Strategy2.R      #   Scenario: Genotype 1,600 seedlings
│   │   ├── Advance_Strategy3.R      #   Scenario: Genotype all F1s
│   │   ├── UpdateParents_Burnin.R   #   Parent selection (burn-in)
│   │   ├── UpdateParents_Scenario.R #   Parent selection (scenarios)
│   │   ├── UpdateResults.R          #   Record metrics per year
│   │   ├── plot_results.R           #   Visualization
│   │   └── rep.txt                  #   SLURM array input
│   │
│   └── SLURM/                       # HPC job submission scripts
│       ├── run_CV1.sbatch
│       ├── run_CV0.sbatch
│       └── run_simulation.sbatch
│
├── .gitignore
└── README.md

---

Analyses

1. CV1 — Within-Environment Prediction (Section 3.2)

Predicts unobserved genotypes within environments where other genotypes have been phenotyped.

• Script: codes/CV1_within_env/GS_CV1_GBLUP.R
• Model: GBLUP (BGLR)
• G matrix: VanRaden for autotetraploids (AGHmatrix, ploidy = 4)
• CV: 5-fold, 10 repetitions per environment × trait
• Traits: Total yield, marketable yield, specific gravity
• Environments: FL_2023, FL_M2_2024, TRS_2023, BG_2023, FL_M1_2024, TRS_2024, BG_2024
• Phenotypes: BLUEs scaled within environment (zero mean, unit variance)

2. CV0 — Across-Environment Prediction (Section 3.3)

Predicts genotype performance in entirely unobserved environments using leave-one-environment-out (LOEO).

• Script: codes/CV0_across_env/GS_CV0_LOEO.R
• Models compared:

Model	Components	Description
M1	G	Genomic kernel only
M2	E + G	Environment main effect + genomic
M3	E + G + G×E	+ Genomic × environment interaction
M4	E + G + G×W	+ Genomic × enviromic interaction

• Environments: FL_2023, FL_M2_2024, TRS_2023, BG_2023, FL_M1_2024, TRS_2024, BG_2024
• Enviromic data: Weather covariance matrix from ECData(VarEnv_raw_allDates_100).RData

3. Platform Comparison (Section 3.4)

Compares prediction accuracy across genotyping platforms and marker densities.

• Script: codes/platform_comparison/GS_CV1_DArT_4K.R — CV1 with 4K markers and Cv1 with imputed set of markers
• Additional scripts for Flex-seq 4K vs 105K and imputation to be added.

4. Breeding Program Simulation (Section 3.5)

Stochastic simulation of a potato breeding pipeline comparing conventional phenotypic selection against genomic selection strategies with varying genotyping intensity.

• Scripts: codes/simulation/ (14 files, orchestrated by RUNME.R)
• Entry point: Rscript RUNME.R <rep> <ddMeanSG>

Pipeline (6 stages per year):

Stage	Description	Reps	Selected
Year 1	Crossing (80 crosses × 200 progeny)	—	16,000
Year 2	Seedling field	—	varies by scenario
Year 3	2nd clonal generation	2	200
Year 4	3rd clonal generation	6	60
Year 5	Official trial 1	15	20
Year 6	Official trial 2	30	3

Scenarios:

Scenario	Description
Conventional	Standard phenotypic selection at all stages
Strategy 1	Genotype 650 seedlings
Strategy 2	Genotype 1,600 seedlings
Strategy 3	Genotype all F1s

Design: 200 tetraploid founders, 12 chromosomes, 300 QTL, 1000 SNPs, 20-year burn-in + 30-year future, 30 replicates via SLURM array.

---

Requirements

• R (≥ 4.x)
• R packages: BGLR, AGHmatrix, AlphaSimR, dplyr, readxl, ggplot2, ggpubr, data.table, plyr

Input Data (to be included)

File	Used by	Description
GS_BLUES_all_FINAL_2.xlsx	CV1, CV0, Platform	BLUEs across environments
Dart_ADM_filtered_4K.csv	CV1, Platform	DArT 4K SNP dosage matrix
M_LGC_DART_4k_ADM_merged.csv	CV0	Merged DArT + LGC SNP dosage matrix
ECData(VarEnv_raw_allDates_100).RData	CV0	Enviromic covariance matrices

Citation

TODO: citation / DOI once published.