Applying multimodal AI to physiological waveforms improves genetic prediction of cardiovascular traits

Abstract

Electronic health records, biobanks, and wearable biosensors enable the collection of multiple health modalities from many individuals. Access to multimodal health data provides a unique opportunity for genetic studies of complex traits because different modalities relevant to a single physiological system (e.g., circulatory system) encode complementary and overlapping information. We propose a multimodal deep learning method, multimodal representation learning for genetic discovery on low-dimensional embeddings (M-REGLE), for discovering genetic associations from a joint representation of complementary electrophysiological waveform modalities. M-REGLE jointly learns a lower representation (i.e., latent factors) of multimodal physiological waveforms using a convolutional variational autoencoder, performs genome-wide association studies (GWASs) on each latent factor, then combines the results to study the genetics of the underlying system. To validate the advantages of M-REGLE and multimodal learning, we apply it to common cardiovascular modalities (photoplethysmogram [PPG] and electrocardiogram [ECG]) and compare its results to unimodal learning methods in which representations are learned from each data modality separately but are statistically combined for downstream genetic comparison. M-REGLE identifies 19.3% more loci on the 12-lead ECG dataset, 13.0% more loci on the ECG lead I + PPG dataset, and its genetic risk score significantly outperforms the unimodal risk score at predicting cardiac phenotypes, such as atrial fibrillation (Afib), in multiple biobanks.

Keywords: AI; GWAS; PRS; cardiovascular disease; deep learning; fusion model; multimodal; representation learning; variational autoencoder.

Graphical abstract

Figure 1

Overview of M-REGLE (A) M-REGLE steps on lead I ECG and PPG, (B) U-REGLE steps on lead I ECG and PPG, (C) M-REGLE steps on 12-lead ECG, and (D) U-REGLE steps on 12-lead ECG. Step 1 in M-REGLE (A and C) obtains the raw embeddings from multimodal health data in a joint fit while step 1 in U-REGLE (B and D) obtains the raw embeddings for each modality separately. In step 2, to ensure completely uncorrelated embeddings, we applied PCA on the raw embeddings. Lastly, we ran GWAS on the uncorrelated embeddings and combined them.

Figure 2

M-REGLE embeddings improve cardiovascular trait prediction (A) Validation reconstruction losses in log scale of U-REGLE and M-REGLE on 12-lead ECG data. The x axis is the numbers of latent dimensions (1, 2, 4, 8, 16, 32, 64) per ECG lead. Standard errors (SEs) are too small to plot (see Table S9 for values of reconstruction losses and SEs. See Figure S25 for the normal scaled reconstruction loss plots.). (B) Validation reconstruction losses in log scale of M-REGLE and U-REGLE across numbers of latent dimensions on lead I ECG and PPG data. The x axis is the numbers of latent dimensions used for each modality (see Table S13 for values of M-REGLE and U-REGLE reconstruction losses and SEs). All the difference between M-REGLE and U-REGLE in (A) and (B) are significant. (C) AUROC prediction of nine phenotypes utilizing ElasticNet trained on the 12 embeddings obtained from ECG lead I and PPG. (D) AUPRC prediction of nine phenotypes utilizing ElasticNet trained on the 12 embeddings obtained from ECG lead I and PPG. Star sign indicates a statistically significant difference between the two methods using paired bootstrapping (100 repetitions) with 95% confidence.

Figure 3

The effect of perturbing distinctive M-REGLE embedding dimensions on decoded ECG and PPG waveforms from a healthy toward an unhealthy sample (A) The impact of the fourth dimension. As the embedding transitions from healthy to unhealthy, the decoded ECG (left) exhibits a prolonged QT interval, while the PPG (right) loses its notch. (B) The impact of the 10th dimension. A shift from healthy to unhealthy along this dimension results in a progressive decrease in ECG amplitude.

Figure 4

M-REGLE on 12 ECG leads increases genomic discovery (A and B) (A) Manhattan plot depicting M-REGLE GWAS p values for all 22 autosomal chromosomes. Black gene names indicate the closest gene for each locus with $-{\log }_{10}p>20$. Purple dots denote the GWS loci uniquely detected by M-REGLE. Orange dots indicate loci also identified in U-REGLE. (B) Comparison of M-REGLE GWS variants in hits with U-REGLE. The x axis is the $-\log$p value of U-REGLE. The y axis is the $-\log$p value of the M-REGLE. All p values in (A) and (B) are computed by summing the chi-squared statistics for all 96 embeddings to perform a single joint chi-squared test. The vertical and horizontal red lines indicate the GWS level. The diagonal red line indicates $y=x$. The orange dots indicate variants in hits that are significant for U-REGLE but not significant for our M-REGLE and green dots indicate variants in hits that are significant for our M-REGLE but not significant for U-REGLE. (C) A three-way Venn diagram of the GWAS Catalog loci, loci discovered by M-REGLE, and loci discovered by U-REGLE. (D) Comparison of the chi-squared statistics for all known significant variants in GWAS Catalog for both U-REGLE and M-REGLE. The difference is statistically significant.

Figure 5

M-REGLE on ECG lead I and PPG increases genomic discovery (A) Manhattan plot depicting M-REGLE GWAS p values. Black gene names indicate the closest gene for each locus with $-{\log }_{10}p>20$. Purple dots denote the GWS loci detected uniquely by M-REGLE. Orange dots indicate loci also identified in U-REGLE. (B) Comparison of M-REGLE GWS variants-in-hits with U-REGLE. The x axis is the $-\log$p value of baseline. The y axis is the $-\log$p value of the M-REGLE. All p values (A) and (B) are computed by summing the chi-squared statistics for all 12 embeddings to perform a single joint chi-squared test. The vertical and horizontal red lines indicate the GWS level. The diagonal red line indicates $y=x$. The orange dots indicate variants in hits that are significant for U-REGLE but not significant for our M-REGLE and green dots indicate variants in hits that are significant for our M-REGLE but not significant for baseline. (C) A three-way Venn diagram of the GWAS Catalog loci, loci discovered by M-REGLE, and loci discovered by U-REGLE. (D) Comparison of the chi-squared statistics for all known significant variants in GWAS Catalog for both U-REGLE and M-REGLE. The difference is statistically significant.

Figure 6

M-REGLE improves Afib genetic risk score (A) The x axis is genetic risk score percentile and y axis is the prevalence. Lower is better for the bottom percentiles; higher is better for the top percentiles. (B and C) (B) AUROC and (C) AUPRC (precision recall) Star sign indicates a statistically significant difference between the two methods using paired bootstrapping (100 repetitions) with 95% confidence.