Skip to content

correlation_analysis_guide.md

Latest commit

 

History

History
73 lines (53 loc) · 2.41 KB

correlation_analysis_guide.md

File metadata and controls

73 lines (53 loc) · 2.41 KB

Cross-Method Correlation Analysis Guide

Overview

This guide describes the gene filtering methodology for improving correlations between MAST (mutations) and CRISPRi (knockdown) results.

Methodology

Traditional Approach

  • Method: Correlate log2FC values for all genes
  • Results: mean |r| = 0.053
  • Issue: Thousands of unchanged genes dilute the signal

Improved Approach

  • Method: Filter to top N most-changed genes per method
  • Results: mean |r| = 0.593 (11x improvement with top 200 genes)
  • Advantage: Focuses on genes most affected by each perturbation

Gene Filtering Options

Top 100 genes

  • Correlation: mean |r| = 0.601
  • Advantage: Highest correlation strength
  • Limitation: Fewer overlapping genes

Top 200 genes (recommended)

  • Correlation: mean |r| = 0.593
  • Advantage: Optimal balance of correlation strength and data points
  • Usage: Default setting in interactive plots

Top 500 genes

  • Correlation: mean |r| = 0.487
  • Advantage: More genes included in analysis
  • Trade-off: Slightly weaker correlations

Implementation

Interactive Plots

Gene filtering is available in the Signature Nomination module:

  1. Navigate to "Gene Pair Analysis"
  2. Select filtering approach from dropdown
  3. View updated correlation plots with trend lines

Analysis Scripts

Correlation analysis scripts are located in:

  • inst/scripts/correlation_analysis/comprehensive_correlation_analysis.R
  • inst/scripts/correlation_analysis/test_correlation_approaches.R

Results

Validation

  • Combinations tested: 180 (12 gene pairs × 3 experiments × 5 clusters)
  • Approaches compared: 7 different filtering strategies
  • Strong correlations: 61 gene pairs with |r| ≥ 0.5

Notable Findings

  • DNAJC6: r = 0.99 (strongest correlation observed)
  • VPS13C variants: Consistent strong correlations across experiments
  • SNCA variants: Strong correlations with directional consistency

Interpretation

Strong correlations indicate biological pathway convergence between:

  • Genetic mutations (MAST analysis)
  • Gene knockdowns (CRISPRi analysis)

This validates that both perturbation methods affect similar downstream pathways, supporting the biological relevance of observed effects.

References

Complete analysis results are available in:

  • inst/results/comprehensive_correlation_results.csv
  • inst/results/correlation_quality_analysis.csv