Skip to content

Single-Input Fine-Mapping

Single-input fine-mapping is the traditional approach where you analyze one cohort or ancestry at a time. This strategy is ideal when you have well-powered individual studies or want to understand ancestry-specific genetic architecture.

When to Use Single-Input Strategy

Single-Input Use Cases

Well-powered individual studies
When each study has sufficient power for fine-mapping alone
Ancestry-specific analysis
When you want to understand population-specific causal variants
Exploratory analysis
When investigating individual study characteristics before meta-analysis
Heterogeneous effects
When effect sizes vary significantly across populations

Step-by-Step Workflow

1. Prepare Single-Study Input

For single-input analysis, your loci file should contain only one row per locus:

single_study_loci.txt
chr start   end popu    sample_size cohort  prefix  locus_id
8   41242482    42492482    EUR 442817  UKBB    data/EUR.UKBB.chr8_41242482_42492482    chr8_41242482_42492482

2. Quality Control

Quality control for single studies focuses on internal consistency:

credtools qc single_study_loci.txt qc_output/

QC Metrics for Single Studies

Inconsistency Parameter (s)
Measures consistency between z-scores and LD matrix
Values > 0.2 suggest potential issues
Kriging RSS
Detects potential allele switches or data quality issues
Flags variants with unexpected z-scores given LD structure
LD Structure Analysis
Eigenvalue decomposition of LD matrix
Identifies problematic LD patterns
MAF Comparison
Compares allele frequencies between summary stats and LD reference
Large differences suggest potential strand issues

Single-Study QC Limitations

Single-study QC cannot detect: - Cross-study heterogeneity
- Population-specific biases - Cohort-specific technical artifacts

Consider multi-study QC when possible.

3. Fine-Mapping

Run fine-mapping on your single study:

credtools finemap single_study_loci.txt output/ \    --tool susie \
    --max-causal 5 \
    --coverage 0.95

Supported Tools for Single-Input

Best for: General purpose, robust across different scenarios

credtools finemap input.txt output/ \    --tool susie \
    --max-causal 10 \
    --max-iter 100 \
    --estimate-residual-variance \
    --convergence-tol 1e-3

Key Parameters:

  • --max-causal: Maximum number of causal variants (default: 1)
  • --max-iter: Maximum iterations (default: 100)
  • --estimate-residual-variance: Estimate phenotype variance (default: False)
  • --convergence-tol: Convergence tolerance (default: 1e-3)

FINEMAP

Best for: Bayesian model averaging, comprehensive uncertainty quantification

credtools finemap input.txt output/ \    --tool finemap \
    --max-causal 5 \
    --n-iter 100000 \
    --n-threads 4

Key Parameters:

  • --n-iter: Number of MCMC iterations (default: 100000)
  • --n-threads: Number of parallel threads (default: 1)

ABF (Approximate Bayes Factors)

Best for: Fast, simple analysis with minimal assumptions

credtools finemap input.txt output/ \    --tool abf \
    --var-prior 0.2

Key Parameters:

  • --var-prior: Prior variance (0.15 for quantitative, 0.2 for binary traits)

CARMA

Best for: Model uncertainty quantification, outlier detection 

credtools finemap input.txt output/ \    --tool carma \
    --max-causal 10 \
    --effect-size-prior "Spike-slab" \
    --y-var 1.0

Key Parameters:

  • --effect-size-prior: "Spike-slab" or "Cauchy" (default: "Spike-slab")
  • --y-var: Phenotype variance (default: 1.0)
  • --bf-threshold: Bayes factor threshold (default: 10.0)

RSparsePro

Best for: Sparse regression approach, computational efficiency

credtools finemap input.txt output/ \    --tool rsparsepro \
    --max-causal 5 \
    --eps 1e-5 \
    --cthres 0.7

Key Parameters:

  • --eps: Convergence criterion (default: 1e-5)
  • --cthres: Coverage threshold (default: 0.7)
  • --minldthres: Minimum LD within effect groups (default: 0.7)

Comparing Single-Input Tools

Tool Speed Memory Uncertainty Best Use Case
SuSiE Fast Low Good General purpose
FINEMAP Moderate Moderate Excellent Comprehensive analysis
ABF Very Fast Very Low Basic Quick screening
CARMA Slow High Excellent Research applications
RSparsePro Fast Low Good Large regions

Automatic Parameter Setting

CREDTOOLS can automatically determine the maximum number of causal variants using COJO:

credtools finemap input.txt output/ \    --tool susie \
    --set-L-by-cojo \
    --p-cutoff 5e-8 \
    --collinear-cutoff 0.9

COJO Parameters:

  • --p-cutoff: P-value threshold for conditioning (default: 5e-8)
  • --collinear-cutoff: Collinearity threshold (default: 0.9)
  • --window-size: Window for conditional analysis (default: 10Mb)
  • --maf-cutoff: MAF cutoff (default: 0.01)

COJO Integration

COJO (Conditional and Joint analysis) estimates the number of independent signals in a region. CREDTOOLS uses this to set --max-causal automatically, which often works better than arbitrary values.

Output Files

Posterior Inclusion Probabilities

output/pips.txt
# SNPID PIP
8-41234567-A-G  0.0234
8-41235678-C-T  0.8765
8-41236789-G-A  0.0456

Credible Sets

output/creds.json
{
  "credible_sets": {
    "cs1": {
      "variants": ["8-41235678-C-T", "8-41235680-A-G"],
      "coverage": 0.95,
      "total_pip": 0.96,
      "min_abs_corr": 0.34
    }
  },
  "tool": "susie",
    "parameters": {...}
}

Common Issues and Solutions

Troubleshooting Single-Input Analysis

No credible sets found
Try increasing --max-causal
Check if region has sufficient signal (--p-cutoff)
Verify LD matrix quality
Very large credible sets
May indicate weak signal or LD issues
Try more stringent --coverage (e.g., 0.99)
Consider region subdivision
Tool convergence issues
Increase --max-iter for iterative methods
Try different --convergence-tol values
Switch to more robust tool (SuSiE)
Memory issues with large regions
Use RSparsePro for efficiency
Consider region subdivision
Reduce LD matrix precision if possible

Example: Complete Single-Study Analysis

# 1. Quality control
credtools qc my_study.txt qc_results/

# 2. Review QC metrics (check s_estimate.txt, kriging_rss.txt)

# 3. Fine-mapping with automatic L setting
credtools finemap my_study.txt results/ \    --tool susie \
    --set-L-by-cojo \
    --coverage 0.95 \
    --max-iter 100

# 4. Review results (pips.txt, creds.json)

Next Steps