Replication-Enhanced Detection of Quantitative Traits under Adaptation

A workflow for detecting adaptive quantitative trait divergence (Q_ST) using Approximate Bayesian Computation (ABC). The workflow supports both local execution (via Snakemake) and distributed computing (via HTCondor).

Workflow Overview

Module 1: Detect Adaptive Traits

1. Calculate variance components using Method of Moments (MoM)
2. Estimate trait Q_ST using ABC with neural network regression
3. Generate neutral Q_ST distribution from empirical F_ST values
4. Compare trait Q_ST to the neutral threshold (95th percentile)
5. Flag traits as adaptive if Q_ST > threshold

Module 2: Evaluate Performance

1. Simulate traits with known adaptive Q_ST values
2. Calculate True Positive Rate (TPR) and False Positive Rate (FPR)
3. Rank summary statistic combinations by performance
4. Compare power across different sample structures

Installation

# Clone the repository
git clone https://github.com/sfeng666/REDQuanTA.git
cd REDQuanTA

# Create conda environment
conda env create -f environment.yml
conda activate redquanta

Quick Start

Module 1: Detect Adaptive Traits (Local/Snakemake)

# Dry run to see what will be executed
snakemake --configfile config/config_detect.yaml --cores 4 -n

# Run the workflow
snakemake --configfile config/config_detect.yaml --cores 4

Module 2: Evaluate Performance (Local/Snakemake)

# With reduced parameters for local testing
snakemake --configfile config/config_evaluate.yaml --cores 4

# Output: results/evaluate/combined_model_ranking.csv

HTCondor Execution

For large-scale analysis using HTCondor distributed computing:

# Module 1: Generate and submit DAG
python htcondor/scripts/prepare_trait_dag.py --trait-id L0MQ04
condor_submit_dag results/dags/trait_L0MQ04.dag

# Module 2: Generate and submit evaluation DAG
python htcondor/scripts/prepare_perf_eval_dag.py --chr both
condor_submit_dag results/perf_eval/perf_eval_autosomes.dag

Input Files

File	Description
sample_structure.csv	Population/strain/replicate structure
trait_values.csv	Trait measurements per strain/replicate
qst_neutral_autosomes.txt	Neutral FST values (autosomes)
qst_neutral_chrX.txt	Neutral FST values (X chromosome)

Sample Structure Format

population,strain,replicate
pop1,strain1,rep1
pop1,strain1,rep2
pop1,strain2,rep1
...

Trait Values Format

trait_id,chr,population,strain,replicate,value
trait001,autosomes,pop1,strain1,rep1,0.523
trait001,autosomes,pop1,strain1,rep2,0.541
...

Output Files

File	Description
qst_results.csv	Detection results (trait_id, chr, QST, adaptive)
tpr_fpr_matrix_*.csv	TPR/FPR matrices per chromosome
combined_model_ranking.csv	Model performance ranking
sample_struct_comparison_*/plots/	Power comparison across sample structures (optional)

For qst_results.csv, blank QST and adaptive fields do not always mean the workflow failed. In some traits, the observed-data ANOVA summary statistics can fall on a boundary case, such as observed QST = 0; then the downstream ABC estimation may return NA/NaN because there is not enough informative variation for a stable posterior estimate. In that case, the threshold columns can still be produced, and the blank estimate should be interpreted as "uninformative for ABC under this sample/trait pattern" rather than as a pipeline error.

Sample Structure Comparison (Optional)

When sample_structures is set in the evaluate config, the workflow runs the evaluation for each structure and produces power comparison plots showing how detection power varies with sample size. By default, the comparison uses the first summary stats combo; set comparison_stats to override.

Directory Structure

REDQuanTA/
├── README.md                    # This file
├── README_details.md            # Detailed documentation
├── environment.yml              # Conda environment
├── Dockerfile                   # Validation container
├── config/                      # Snakemake configuration
│   ├── config_detect.yaml       # Module 1 config
│   ├── config_evaluate.yaml     # Module 2 (local, reduced params)
│   └── config_evaluate_full.yaml# Module 2 (HTCondor, full params)
├── workflow/                    # Snakemake workflow
│   ├── Snakefile
│   ├── rules/
│   └── scripts/
├── htcondor/                    # HTCondor execution
│   ├── scripts/
│   └── env/
├── data/
│   ├── example/                 # Example input data
│   └── reference/               # Reference figures
└── results/                     # Output directory

Documentation

• README_details.md: Full parameter documentation, HTCondor setup, troubleshooting
• HTCondor Documentation: Official HTCondor docs

Citation

If you use REDQuanTA in your research, please cite:

Feng, S., & Pool, J. E. (in preparation). REDQuanTA: Replication-Enhanced Detection of Quantitative Traits under Adaptation — an improved statistical framework to detect locally adaptive traits. Laboratory of Genetics, University of Wisconsin-Madison.

Until the paper is published, you may also cite this repository directly:

Feng, S., & Pool, J. E. (2025). REDQuanTA [Software]. GitHub. https://github.com/sfeng666/REDQuanTA

License

This project is licensed under the MIT License - see the LICENSE file for details.