Unsupervised representation learning improves genomic discovery and risk prediction for respiratory and circulatory functions and diseases

Abstract

Background: High-dimensional clinical data are becoming more accessible in biobank-scale datasets. However, accurately phenotyping high-dimensional clinical data remains a major impediment to genetic discovery. Methods: We introduce a general deep learning framework, REpresentation learning for Genetic discovery on Low-dimensional Embeddings (REGLE), for discovering associations between genetic variants and high-dimensional clinical data. REGLE uses convolutional variational autoencoders to compute a non-linear, low-dimensional, disentangled embedding of the data and can also incorporate expert clinical metrics. We demonstrate the utility of REGLE by application to spirograms, which measure lung function. We generate two types of synthetic representations of pulmonary functions we call spirogram encodings (SPINCs) and residual spirogram encodings (RSPINCs). Findings: Genome-wide association studies on (R)SPINCs identify more genome-wide significant loci than existing methods while replicating most known lung function loci. Furthermore, (R)SPINCs are associated with overall survival and, under the latent causal variable model, they exhibit significantly high genetic causality proportion with asthma, chronic obstructive pulmonary disease (COPD), and inflammatory diseases. Finally, we construct a set of polygenic risk scores (PRS) that are generally predictive of pulmonary traits and diseases. We demonstrate superior performance predicting asthma and COPD, in multiple ancestries and across four biobanks, compared to PRSs constructed using expert-defined pulmonary function measurements. Interpretation: REGLE is a method for generating low-dimensional, disentangled representations of high-dimensional clinical data that does not require labels, and improves upon expert-defined phenotypes for genetic discovery and disease prediction. It can flexibly incorporate expert-defined or clinical features and provides a framework to create accurate disease-specific PRS in datasets which have minimal expert phenotyping. (R)SPINCs are quantifying clinically relevant features that are not currently captured in a standardized or automated way.

Utilizing multimodal AI to improve genetic analyses of cardiovascular traits

“…Next, we analyzed ECG lead I plus PPG and observed a modest correlation, indicative of complementary signals (0.43 ± 0.01) (Supplementary Table 2). As a reference point, we observed that the spirogram data–a measure of lung function–in UKB [7, 8] has a lower projected correlation with lead I ECG (0.29 ± 0.03) and PPG (0.37 ± 0.03). It is worth noting that we do expect to see some non-zero correlation as HDCDs capture general health information of individuals such as age, sex, and BMI.…”

“…However, we currently lack statistical methods to fully utilize these multimodal HDCD in genetic analyses. REGLE (Yun et al) [8] provides a means to study the genetic underpinnings of high dimensional data, but it is limited to one data modality, and does not leverage the shared and complementary information in multimodal HDCD. Here, we developed an unsupervised representation learning method, M-REGLE, which aims to utilize multimodal HDCD in a joint model to improve genetic analyses.…”

“…Then preliminary GWASes were performed on each of the PCs of the representations for each data setting and modality type using REGENIE [28]. The GWASes were adjusted for age, sex, body mass index, smoking status, genotyping array and the top 15 genetic PCs [7, 8].…”

“…REGLE uses convolutional variational autoencoders (VAEs) to compute a non-linear, low-dimensional, disentangled embedding of the data without clinical labels. Although REGLE’s performance in genetic analysis and down-stream polygenic risk scoring (PRS) surpasses prior methods [8], the utility of REGLE is limited to one data modality.…”

Deep Learning Utilizing Suboptimal Spirometry Data to Improve Lung Function and Mortality Prediction in the UK Biobank

“…In addition, the neural network parametrization of the Spiro-CLF model encodes nonlinear effects from the raw spirometry efforts. Previous works have shown that using neural networks, especially in conjunction with convolution layers, can improve representations of lung function with respect to prediction of COPD (Bhattacharjee et al, 2022), COPD subtypes (Bodduluri et al, 2020), and gene association Yun et al, 2023).…”