Wavelet Decomposition-Based Genomic Analysis of the Human Electrocardiogram

Figures

• 

  Figure 1.. Wavelet basis functions for Haar and Daubechies-6 (db6). Each wavelet family is characterized by a scaling function φ (blue, solid) and a wavelet function ψ (red, dashed). These shapes are not chosen arbitrarily; they are the geometric fingerprint of the filter coefficients: φ emerges from repeatedly applying the low-pass filter h to itself, and ψ from the complementary high-pass filter g thus φ and ψ reveal what each filter is sensitive to. Left: the Haar scaling function is a flat step, discontinuous by construction, producing blocky artifacts when applied to smooth cardiac signals. Right: db6’s scaling function exhibits a sharp central peak with smooth, decaying oscillations that closely resemble the QRS complex morphology, making it geometrically well-suited to capture heartbeat energy in the approximation coefficients. X-axis: time; Y-axis: amplitude.

• 

  Figure 2.. Wavelet decomposition of a single-lead ECG signal and overview of electrodes placements. The ECG signal (Lead I) is decomposed using the Daubechies-6 (db6) wavelet at 6 levels. Column 1 shows the original ECG and the standard electrode configuration used to record 12-lead ECGs. Column 2 displays the reconstructed signal of each individual level (A6, D6–D1), obtained by inverse wavelet transform with all other coefficients set to zero, and including a zoomed inset highlighting a 2-second segment. Column 3 illustrates reconstruction of the signal by cumulatively adding wavelet levels starting from the approximation A6 (lowest frequency) and progressively including detail coefficients D6 to D1 (increasing frequencies). X-axis: time; Y-axis: amplitude.

• 

  Figure 3.. Heatmap of energy features across ECG leads and decomposition level. X-axis: decomposition levels; Y-axis: 12 leads. Each row represents a lead and each column represents a decomposition level (detail or approximation). The color intensity corresponds to the magnitude of the energy at that lead and level.

• 

  Figure 4A.. Manhattan plot displaying genome-wide association results aggregated across 84 waveform energy phenotypes (12 ECG leads × 7 wavelet decomposition levels), each analyzed in a separate GWAS. Each point represents a SNP plotted by chromosomal position (x-axis) against its association significance expressed as −log10(p-value) (y-axis). P-values were derived from ordinary least squares linear regression (PLINK2), testing the additive effect of allele dosage on quantile-normalized waveform energy at each lead–level combination, with adjustment for age, sex, and 10 genetic principal components. Horizontal lines indicate suggestive (p = 1×10−5, red) and genome-wide significance thresholds (p = 5×10−8, green).

• 

  Figure 4B.. Manhattan plot of 67 independent lead SNPs identified after LD clumping (r2 < 0.1, 500 kb window) applied to each of the 84 GWAS separately. Each point represents the index variant with the strongest association within each correlated SNP set, plotted by chromosomal position (x-axis) against −log10(p-value) (y-axis). SNPs are colored by significance tier: red for variants surpassing the Bonferroni-corrected threshold (p = 1.18 × 10−8), and green for variants meeting the conventional genome-wide significance threshold (p = 5 × 10−8).

• 

  Figure 5.. Heatmap of the heritability (h2) estimates of energy features across ECG leads and decomposition level. X-axis: decomposition levels; Y-axis: 12 leads. Each cell shows the SNP-heritability estimated using LD Score Regression (--h2) with European LD scores from the 1000 Genomes Project [13]. For each lead-level energy phenotype, GWAS summary statistics were regressed on LD scores, and h2 was computed from the slope of this regression.

• 

  Figure 6.. Heatmap of genetic correlations (rg) for energy features between Leads I and aVR across wavelet decomposition levels. X-axis: energy levels for aVR lead; Y-axis: energy levels for Lead I. Each cell shows the genetic correlation estimated using LD Score Regression (--rg) with European LD scores from the 1000 Genomes Project reference panel. For each pair of lead–level energy phenotypes, GWAS summary statistics were jointly analyzed by regressing the product of SNP z-scores across traits on LD scores, and the genetic correlation (rg) was derived from the slope of this cross-trait regression, scaled by the square roots of the corresponding SNP-heritability estimates.

• 

  Figure 7.. Clustered bubble plot of genetic correlations (rg) between 84 ECG energy features (12 leads × 6 detail levels + 1 approximation level) and 7 FinnGen cardiovascular phenotypes. Each cell shows the genetic correlation estimated using LD Score Regression. X-axis: energy features; Y-axis: FinnGenn diseases. Only associations with p < 0.005 are shown. Larger and darker bubbles indicate stronger genetic correlations. Lead and level combinations such as Lead I D6 and V1 D5 exhibit the strongest associations, particularly for heart failure and coronary heart disease. These spatial patterns reflect disease-specific electrophysiological and genetic signatures across the 12-lead ECG.

• 

  Figure 8.. Gene enrichment analysis of 110 genes mapped from the clumped leading SNPs via FUMA snps2genes, submitted to Enrichr against the Jensen DISEASES Curated 2025 (ontology enrichment, left) and CellMarker 2024 (cell-type enrichment, right) libraries. X-axis: enriched terms; Y-axis: genes. For each term, enrichment significance was assessed by testing whether the overlap between submitted genes and the term’s gene set exceeded what would be expected by chance. Color intensity reflects –−log10(p-value), with darker red indicating greater statistical enrichment. Left: enriched disease ontology terms highlight convergent genetic associations with cardiomyopathy, arrhythmia, and related cardiac pathologies. Right: cell-type enrichment reveals preferential expression of associated genes in cardiomyocytes, consistent with a cardiac-specific genetic architecture.

• 

  Figure 9.. Heatmap of genetic correlations (rg) between 84 ECG energy features (12 leads × 6 detail levels + 1 approximation level) and BMI. Each cell shows the genetic correlation estimated using LD Score Regression. X-axis: decomposition levels (A6, D6–D1); Y-axis: 12 ECG leads. Color intensity reflects the magnitude and direction of rg, with red indicating positive correlations and blue indicating negative correlations.

• 

  Figure 10.. Forest plot of regression coefficients (β) for the 84 ECG wavelet energy features on five standard UK Biobank machine-reported ECG metrics (QRS duration, QT interval, P-wave duration, heart rate, R-wave axis). Each cell shows β as a Gaussian density centred on the point estimate with width proportional to the standard error; distributions not overlapping zero indicate effects reliably distinct from zero. X-axis: β. Y-axis: 12 ECG leads, each containing 7 distributions for wavelet levels A6, D6–D1. Colour intensity reflects −log10(p-value), with bright yellow indicating high significance and dark blue weak association.