Sex-biased Regulation of Extracellular Matrix Genes in COPD.

Compared to men, women often develop COPD at an earlier age with worse respiratory symptoms despite lower smoking exposure. However, most preventive, and therapeutic strategies ignore biological sex differences in COPD. Our goal was to better understand sex-specific gene regulatory processes in lung tissue and the molecular basis for sex differences in COPD onset and severity. We analyzed lung tissue gene expression and DNA methylation data from 747 individuals in the Lung Tissue Research Consortium (LTRC), and 85 individuals in an independent dataset. We identified sex differences in COPD-associated gene regulation using gene regulatory networks. We used linear regression to test for sex-biased associations of methylation with lung function, emphysema, smoking, and age. Analyzing gene regulatory networks in the control group, we identified that genes involved in the extracellular matrix (ECM) have higher transcriptional factor targeting in females than in males. However, this pattern is reversed in COPD, with males showing stronger regulatory targeting of ECM-related genes than females. Smoking exposure, age, lung function, and emphysema were all associated with sex-specific differential methylation of ECM-related genes. We identified sex-based gene regulatory patterns of ECM-related genes associated with lung function and emphysema. Multiple factors including epigenetics, smoking, aging, and cell heterogeneity influence sex-specific gene regulation in COPD. Our findings underscore the importance of considering sex as a key factor in disease susceptibility and severity.

COPD Subtypes Are Differentially Associated with Cardiovascular Events and COPD Exacerbations.

BACKGROUND: The coronary artery calcium score (CACS) and ratio of the pulmonary artery to aorta diameters (PA:A ratio) measured from chest CT scans have been established as predictors of cardiovascular events and chronic obstructive pulmonary disease (COPD) exacerbations, respectively. However, little is known about the reciprocal relationship between these predictors and outcomes. Furthermore, the prognostic implications of COPD subtypes on clinical outcomes remain insufficiently characterized. RESEARCH QUESTION: How can these two chest CT-derived parameters predict subsequent cardiovascular events and COPD exacerbations in different COPD subtypes? STUDY DESIGN AND METHODS: Using COPDGene study data, we assessed prospective cardiovascular disease (CVD) and COPD exacerbation risk in COPD subjects (Global Initiative for Chronic Obstructive Lung Disease spirometric grades 2-4), focusing on CACS and PA:A ratio at study enrollment, with logistic regression models. These outcomes were analyzed in three COPD subtypes: 1,042 Non-emphysema-predominant COPD (NEPD; low attenuation area at -950 Hounsfield units [LAA-950]<5%), 1,324 Emphysema-predominant COPD (EPD; LAA-950≥10%), and 465 Intermediate Emphysema COPD (IE; 5≤LAA-950<10%). RESULTS: Our study indicated significantly higher overall risk for cardiovascular events in subjects with higher CACS (≥median; Odds Ratio (OR): 1.61, 95% Confidence Interval (CI)=1.30-2.00) and increased COPD exacerbations in those with higher PA:A ratios (≥1; OR: 1.80, 95% CI=1.46-2.23). Notably, NEPD subjects showed a stronger association between these indicators and clinical events compared to EPD (with CACS/CVD, NEPD vs. EPD, OR 2.02 vs. 1.41; with PA:A ratio/COPD exacerbation, NEPD vs. EPD, OR 2.50 vs. 1.65); the difference in odds ratios between COPD subtypes was statistically significant for CACS/CVD. INTERPRETATION: Two chest CT parameters, CACS and PA:A ratio, hold distinct predictive values for cardiovascular events and COPD exacerbations that are influenced by specific COPD subtypes. TRIAL REGISTRATION: ClinicalTrials.gov Identifier: NCT00608764.

Overlap between COPD genetic association results and transcriptional quantitative trait loci.

Rationale: Genome-wide association studies (GWAS) have identified multiple genetic loci associated with chronic obstructive pulmonary disease (COPD). When integrated with GWAS results, expression quantitative trait locus (eQTL) studies can provide insight into biological mechanisms involved in disease by identifying single nucleotide polymorphisms (SNPs) that contribute to whole gene expression. However, there are multiple genetically driven regulatory and isoform-specific effects which cannot be detected in traditional eQTL analyses. Here, we identify SNPs that are associated with alternative splicing (sQTL) in addition to eQTLs to identify novel functions for COPD associated genetic variants. Methods: We performed RNA sequencing on whole blood from 3743 subjects in the COPDGene Study. RNA sequencing data from lung tissue of 1241 subjects from the Lung Tissue Research Consortium (LTRC), and whole genome sequencing data on all subjects. Associations between all SNPs within 1000 kb of a gene (cis-) and splice and gene expression quantifications were tested using tensorQTL. In COPDGene a total of 11,869,333 SNPs were tested for association with 58,318 splice clusters, and 8,792,206 SNPs were tested for association with 70,094 splice clusters in LTRC. We assessed colocalization with COPD-associated SNPs from a published GWAS[1]. Results: After adjustment for multiple statistical testing, we identified 28,110 splice-sites corresponding to 3,889 unique genes that were significantly associated with genotype in COPDGene whole blood, and 58,258 splice-sites corresponding to 10,307 unique genes associated with genotype in LTRC lung tissue. We found 7,576 sQTL splice-sites corresponding to 2,110 sQTL genes were shared between whole blood and lung, while 20,534 sQTL splice-sites in 3,518 genes were unique to blood and 50,682 splice-sites in 9,677 genes were unique to lung. To determine what proportion of COPD-associated SNPs were associated with transcriptional splicing, we performed colocalization analysis between COPD GWAS and sQTL data, and found that 38 genomic windows, corresponding to 38 COPD GWAS loci had evidence of colocalization between QTLs and COPD. The top five colocalizations between COPD and lung sQTLs include NPNT , FBXO38 , HHIP , NTN4 and BTC . Conclusions: A total of 38 COPD GWAS loci contain evidence of sQTLs, suggesting that analysis of sQTLs in whole blood and lung tissue can provide novel insights into disease mechanisms.

Leukemia inhibitory factor (LIF) receptor amplifies pathogenic activation of fibroblasts in lung fibrosis.

Fibrosis drives end-organ damage in many diseases. However, clinical trials targeting individual upstream activators of fibroblasts, such as TGFß, have largely failed. Here, we target the leukemia inhibitory factor receptor (LIFR) as a "master amplifier" of multiple upstream activators of lung fibroblasts. In idiopathic pulmonary fibrosis (IPF), the most common fibrotic lung disease, we found that lung myofibroblasts had high LIF expression. Further, TGFß1, one of the key drivers of fibrosis, upregulated LIF expression in IPF fibroblasts. In vitro anti-LIFR antibody blocking on human IPF lung fibroblasts reduced induction of profibrotic genes downstream of TGFß1, IL-4 and IL-13. Further, siRNA silencing of LIFR in IPF precision cut lung slices reduced expression of fibrotic proteins. Together, we find that LIFR drives an autocrine positive feedback loop that amplifies and sustains pathogenic activation of IPF fibroblasts downstream of multiple external stimuli, implicating LIFR as a therapeutic target in fibrosis. Significance Statement: Fibroblasts have a central role in the pathogenesis of fibrotic diseases. However, due to in part to multiple profibrotic stimuli, targeting a single activator of fibroblasts, like TGFß, has not yielded successful clinical treatments. We hypothesized that a more effective therapeutic strategy is identifying a downstream "master amplifier" of a range of upstream profibrotic stimuli. This study identifies the leukemia inhibitory factor receptor (LIFR) on fibrotic lung fibroblasts amplifies multiple profibrotic stimuli, such as IL-13 and TGFß. Blocking LIFR reduced fibrosis in ex vivo lung tissue from patients with idiopathic pulmonary fibrosis (IPF). LIFR, acting as a master amplifier downstream of fibroblast activation, offers an alternative therapeutic strategy for fibrotic diseases.

Polygenic and transcriptional risk scores identify chronic obstructive pulmonary disease subtypes.

Rationale: Genetic variants and gene expression predict risk of chronic obstructive pulmonary disease (COPD), but their effect on COPD heterogeneity is unclear. Objectives: Define high-risk COPD subtypes using both genetics (polygenic risk score, PRS) and blood gene expression (transcriptional risk score, TRS) and assess differences in clinical and molecular characteristics. Methods: We defined high-risk groups based on PRS and TRS quantiles by maximizing differences in protein biomarkers in a COPDGene training set and identified these groups in COPDGene and ECLIPSE test sets. We tested multivariable associations of subgroups with clinical outcomes and compared protein-protein interaction networks and drug repurposing analyses between high-risk groups. Measurements and Main Results: We examined two high-risk omics-defined groups in non-overlapping test sets (n=1,133 NHW COPDGene, n=299 African American (AA) COPDGene, n=468 ECLIPSE). We defined "High activity" (low PRS/high TRS) and "severe risk" (high PRS/high TRS) subgroups. Participants in both subgroups had lower body-mass index (BMI), lower lung function, and alterations in metabolic, growth, and immune signaling processes compared to a low-risk (low PRS, low TRS) reference subgroup. "High activity" but not "severe risk" participants had greater prospective FEV 1 decline (COPDGene: -51 mL/year; ECLIPSE: - 40 mL/year) and their proteomic profiles were enriched in gene sets perturbed by treatment with 5-lipoxygenase inhibitors and angiotensin-converting enzyme (ACE) inhibitors. Conclusions: Concomitant use of polygenic and transcriptional risk scores identified clinical and molecular heterogeneity amongst high-risk individuals. Proteomic and drug repurposing analysis identified subtype-specific enrichment for therapies and suggest prior drug repurposing failures may be explained by patient selection.

Lung Transcriptomics Links Emphysema to Barrier Dysfunction and Macrophage Subpopulations.

RATIONALE: While many studies have examined gene expression in lung tissue, the gene regulatory processes underlying emphysema are still not well understood. Finding efficient non-imaging screening methods and disease-modifying therapies has been challenging, but knowledge of the transcriptomic features of emphysema may help in this effort. OBJECTIVES: Our goals were to identify emphysema-associated biological pathways through transcriptomic analysis of bulk lung tissue, to determine the lung cell types in which these emphysema-associated pathways are altered, and to detect unique and overlapping transcriptomic signatures in blood and lung samples. METHODS: Using RNA-sequencing data from 446 samples in the Lung Tissue Research Consortium (LTRC) and 3,606 blood samples from the COPDGene study, we examined the transcriptomic features of chest computed tomography-quantified emphysema. We also leveraged publicly available lung single-cell RNA-sequencing data to identify cell types showing COPD-associated differential expression of the emphysema pathways found in the bulk analyses. MEASUREMENTS AND MAIN RESULTS: In the bulk lung RNA-seq analysis, 1,087 differentially expressed genes and 34 dysregulated pathways were significantly associated with emphysema. We observed alternative splicing of several genes and increased activity in pluripotency and cell barrier function pathways. Lung tissue and blood samples shared differentially expressed genes and biological pathways. Multiple lung cell types displayed dysregulation of epithelial barrier function pathways, and distinct pathway activities were observed among various macrophage subpopulations. CONCLUSIONS: This study identified emphysema-related changes in gene expression and alternative splicing, cell-type specific dysregulated pathways, and instances of shared pathway dysregulation between blood and lung.

Correlation-based network integration of lung RNA sequencing and DNA methylation data in chronic obstructive pulmonary disease.

Chronic Obstructive Pulmonary Disease (COPD) is a heterogeneous, chronic inflammatory process of the lungs and, like other complex diseases, is caused by both genetic and environmental factors. Detailed understanding of the molecular mechanisms of complex diseases requires the study of the interplay among different biomolecular layers, and thus the integration of different omics data types. In this study, we investigated COPD-associated molecular mechanisms through a correlation-based network integration of lung tissue RNA-seq and DNA methylation data of COPD cases (n = 446) and controls (n = 346) derived from the Lung Tissue Research Consortium. First, we performed a SWIM-network based analysis to build separate correlation networks for RNA-seq and DNA methylation data for our case-control study population. Then, we developed a method to integrate the results into a coupled network of differentially expressed and differentially methylated genes to investigate their relationships across both molecular layers. The functional enrichment analysis of the nodes of the coupled network revealed a strikingly significant enrichment in Immune System components, both innate and adaptive, as well as immune-system component communication (interleukin and cytokine-cytokine signaling). Our analysis allowed us to reveal novel putative COPD-associated genes and to analyze their relationships, both at the transcriptomics and epigenomics levels, thus contributing to an improved understanding of COPD pathogenesis.

Machine learning models for predicting blood pressure phenotypes by combining multiple polygenic risk scores.

We construct non-linear machine learning (ML) prediction models for systolic and diastolic blood pressure (SBP, DBP) using demographic and clinical variables and polygenic risk scores (PRSs). We developed a two-model ensemble, consisting of a baseline model, where prediction is based on demographic and clinical variables only, and a genetic model, where we also include PRSs. We evaluate the use of a linear versus a non-linear model at both the baseline and the genetic model levels and assess the improvement in performance when incorporating multiple PRSs. We report the ensemble model's performance as percentage variance explained (PVE) on a held-out test dataset. A non-linear baseline model improved the PVEs from 28.1 to 30.1% (SBP) and 14.3% to 17.4% (DBP) compared with a linear baseline model. Including seven PRSs in the genetic model computed based on the largest available GWAS of SBP/DBP improved the genetic model PVE from 4.8 to 5.1% (SBP) and 4.7 to 5% (DBP) compared to using a single PRS. Adding additional 14 PRSs computed based on two independent GWASs further increased the genetic model PVE to 6.3% (SBP) and 5.7% (DBP). PVE differed across self-reported race/ethnicity groups, with primarily all non-White groups benefitting from the inclusion of additional PRSs. In summary, non-linear ML models improves BP prediction in models incorporating diverse populations.

Biosurfer for systematic tracking of regulatory mechanisms leading to protein isoform diversity.

Long-read RNA sequencing has shed light on transcriptomic complexity, but questions remain about the functionality of downstream protein products. We introduce Biosurfer, a computational approach for comparing protein isoforms, while systematically tracking the transcriptional, splicing, and translational variations that underlie differences in the sequences of the protein products. Using Biosurfer, we analyzed the differences in 32,799 pairs of GENCODE annotated protein isoforms, finding a majority (70%) of variable N-termini are due to the alternative transcription start sites, while only 9% arise from 5' UTR alternative splicing. Biosurfer's detailed tracking of nucleotide-to-residue relationships helped reveal an uncommonly tracked source of single amino acid residue changes arising from the codon splits at junctions. For 17% of internal sequence changes, such split codon patterns lead to single residue differences, termed "ragged codons". Of variable C-termini, 72% involve splice- or intron retention-induced reading frameshifts. We found an unusual pattern of reading frame changes, in which the first frameshift is closely followed by a distinct second frameshift that restores the original frame, which we term a "snapback" frameshift. We analyzed long read RNA-seq-predicted proteome of a human cell line and found similar trends as compared to our GENCODE analysis, with the exception of a higher proportion of isoforms predicted to undergo nonsense-mediated decay. Biosurfer's comprehensive characterization of long-read RNA-seq datasets should accelerate insights of the functional role of protein isoforms, providing mechanistic explanation of the origins of the proteomic diversity driven by the alternative splicing. Biosurfer is available as a Python package at https://github.com/sheynkman-lab/biosurfer.

Computational Deconvolution of Cell Type-Specific Gene Expression in COPD and IPF Lungs Reveals Disease Severity Associations.

RATIONALE: Chronic obstructive pulmonary disease (COPD) and idiopathic pulmonary fibrosis (IPF) are debilitating diseases associated with divergent histopathological changes in the lungs. At present, due to cost and technical limitations, profiling cell types is not practical in large epidemiology cohorts (n>1000). Here, we used computational deconvolution to identify cell types in COPD and IPF lungs whose abundances and cell type-specific gene expression are associated with disease diagnosis and severity. METHODS: We analyzed lung tissue RNA-seq data from 1026 subjects (COPD, n=465; IPF, n=213; control, n=348) from the Lung Tissue Research Consortium. We performed RNA-seq deconvolution, querying thirty-eight discrete cell-type varieties in the lungs. We tested whether deconvoluted cell-type abundance and cell type-specific gene expression were associated with disease severity. RESULTS: The abundance score of twenty cell types significantly differed between IPF and control lungs. In IPF subjects, eleven and nine cell types were significantly associated with forced vital capacity (FVC) and diffusing capacity for carbon monoxide (DLCO), respectively. Aberrant basaloid cells, a rare cells found in fibrotic lungs, were associated with worse FVC and DLCO in IPF subjects, indicating that this aberrant epithelial population increased with disease severity. Alveolar type 1 and vascular endothelial (VE) capillary A were decreased in COPD lungs compared to controls. An increase in macrophages and classical monocytes was associated with lower DLCO in IPF and COPD subjects. In both diseases, lower non-classical monocytes and VE capillary A cells were associated with increased disease severity. Alveolar type 2 cells and alveolar macrophages had the highest number of genes with cell type-specific differential expression by disease severity in COPD and IPF. In IPF, genes implicated in the pathogenesis of IPF, such as matrix metallopeptidase 7, growth differentiation factor 15, and eph receptor B2, were associated with disease severity in a cell type-specific manner. CONCLUSION: Utilization of RNA-seq deconvolution enabled us to pinpoint cell types present in the lungs that are associated with the severity of COPD and IPF. This knowledge offers valuable insight into the alterations within tissues in more advanced illness, ultimately providing a better understanding of the underlying pathological processes that drive disease progression.

HHIP protein interactions in lung cells provide insight into COPD pathogenesis.

Chronic obstructive pulmonary disease (COPD) is the third leading cause of death worldwide. The primary causes of COPD are environmental, including cigarette smoking; however, genetic susceptibility also contributes to COPD risk. Genome-Wide Association Studies (GWASes) have revealed more than 80 genetic loci associated with COPD, leading to the identification of multiple COPD GWAS genes. However, the biological relationships between the identified COPD susceptibility genes are largely unknown. Genes associated with a complex disease are often in close network proximity, i.e. their protein products often interact directly with each other and/or similar proteins. In this study, we use affinity purification mass spectrometry (AP-MS) to identify protein interactions with HHIP , a well-established COPD GWAS gene which is part of the sonic hedgehog pathway, in two disease-relevant lung cell lines (IMR90 and 16HBE). To better understand the network neighborhood of HHIP , its proximity to the protein products of other COPD GWAS genes, and its functional role in COPD pathogenesis, we create HUBRIS, a protein-protein interaction network compiled from 8 publicly available databases. We identified both common and cell type-specific protein-protein interactors of HHIP. We find that our newly identified interactions shorten the network distance between HHIP and the protein products of several COPD GWAS genes, including DSP, MFAP2, TET2 , and FBLN5 . These new shorter paths include proteins that are encoded by genes involved in extracellular matrix and tissue organization. We found and validated interactions to proteins that provide new insights into COPD pathobiology, including CAVIN1 (IMR90) and TP53 (16HBE). The newly discovered HHIP interactions with CAVIN1 and TP53 implicate HHIP in response to oxidative stress.

Colocalization analysis of 3' UTR alternative polyadenylation quantitative trait loci reveals novel mechanisms underlying associations with lung function.

While many disease-associated single nucleotide polymorphisms (SNPs) are expression quantitative trait loci (eQTLs), a large proportion of genome-wide association study (GWAS) variants are of unknown function. Alternative polyadenylation (APA) plays an important role in posttranscriptional regulation by allowing genes to shorten or extend 3' untranslated regions (UTRs). We hypothesized that genetic variants that affect APA in lung tissue may lend insight into the function of respiratory associated GWAS loci. We generated alternative polyadenylation (apa) QTLs using RNA sequencing and whole genome sequencing on 1241 subjects from the Lung Tissue Research Consortium (LTRC) as part of the NHLBI TOPMed project. We identified 56 179 APA sites corresponding to 13 582 unique genes after filtering out APA sites with low usage. We found that a total of 8831 APA sites were associated with at least one SNP with q-value < 0.05. The genomic distribution of lead APA SNPs indicated that the majority are intronic variants (33%), followed by downstream gene variants (26%), 3' UTR variants (17%), and upstream gene variants (within 1 kb region upstream of transcriptional start site, 10%). APA sites in 193 genes colocalized with GWAS data for at least one phenotype. Genes containing the top APA sites associated with GWAS variants include membrane associated ring-CH-type finger 2 (MARCHF2), nectin cell adhesion molecule 2 (NECTIN2), and butyrophilin subfamily 3 member A2 (BTN3A2). Overall, these findings suggest that APA may be an important mechanism for genetic variants in lung function and chronic obstructive pulmonary disease (COPD).

Predictors of Nicotine Replacement Therapy Adherence: Mixed-Methods Research With a Convergent Parallel Design.

BACKGROUND: Few studies have examined the effect of baseline attitudes toward nicotine replacement therapy (NRT) on its actual adherence in a smoking cessation intervention. PURPOSE: This study (i) examined the predictability of baseline variables (quantitative data) on NRT adherence and (ii) explored the congruence of participants' statements about NRT products (qualitative data) during counseling sessions with their baseline attitudes. METHODS: This is a mixed-methods research study using a convergent parallel design. Participants included 74 individuals in the treatment group who received behavioral counseling and combination NRT. A Poisson regression analysis was performed to identify baseline variables predicting NRT adherence. Thematic analysis was completed with a subset of participants (n = 38) who varied in NRT attitude scores and adherence. A joint display was created to integrate quantitative and qualitative data and discover convergence. RESULTS: Approximately 59% of the participants (41/74) used NRT continuously for ≥5 weeks. Having negative attitudes toward NRT and depressive symptoms predicted NRT adherence even after controlling for education and anxiety symptoms. Thematic analysis revealed that NRT adherence is a learning process that consists of the following three distinctive but interrelated phases: (i) information needs, (ii) comprehensive readiness, and (iii) experiential learning. Of the 38 participants, 34 (89.5%) showed convergence between baseline attitude scores and statements about NRT made during counseling sessions. CONCLUSIONS: Individuals who have negative attitudes toward NRT are less likely to use the products in a smoking cessation intervention. Counselors should assess attitudes toward NRT at baseline and address them proactively during counseling sessions.

Few research studies have explored how attitudes toward nicotine substitutes (nicotine patches, gum, and lozenges) affect people's adherence to those substitutes (using them consistently as directed). This study examined (i) whether age, gender, education, attitudes toward the substitutes, and depressive and anxiety symptoms would predict peoples' adherence to these nicotine substitutes during a study to help stop smoking and (ii) whether peoples' statements about their experiences with the substitutes would reveal any patterns. The study was conducted with 74 individuals who received behavioral counseling and combination nicotine substitutes. Having negative attitudes toward the substitutes and depressive symptoms predicted adherence. Age, gender, education, positive attitudes, and anxiety symptoms did not. Statements from a subset of participants (n = 38) revealed that adherence to the substitutes is a learning process that consists of the following three phases: (i) needing more information assuring the safety of the substitutes, (ii) being mentally and situationally ready, and (iii) learning while being involved in the process such as "trial and error." Individuals who have negative attitudes toward the substitutes are less likely to use them, and counselors should assess attitudes toward nicotine replacement therapy before suggesting their use and address these attitudes proactively during smoking cessation counseling sessions.

A Generalized Higher-order Correlation Analysis Framework for Multi-Omics Network Inference.

Multiple -omics (genomics, proteomics, etc.) profiles are commonly generated to gain insight into a disease or physiological system. Constructing multi-omics networks with respect to the trait(s) of interest provides an opportunity to understand relationships between molecular features but integration is challenging due to multiple data sets with high dimensionality. One approach is to use canonical correlation to integrate one or two omics types and a single trait of interest. However, these types of methods may be limited due to (1) not accounting for higher-order correlations existing among features, (2) computational inefficiency when extending to more than two omics data when using a penalty term-based sparsity method, and (3) lack of flexibility for focusing on specific correlations (e.g., omics-to-phenotype correlation versus omics-to-omics correlations). In this work, we have developed a novel multi-omics network analysis pipeline called Sparse Generalized Tensor Canonical Correlation Analysis Network Inference (SGTCCA-Net) that can effectively overcome these limitations. We also introduce an implementation to improve the summarization of networks for downstream analyses. Simulation and real-data experiments demonstrate the effectiveness of our novel method for inferring omics networks and features of interest.

Disentangling Predictors of COPD Mortality with Probabilistic Graphical Models.

Background-Research question: Chronic Obstructive Pulmonary Disease (COPD) is a leading cause of mortality. Predicting mortality risk in COPD patients can be important for disease management strategies. Although scores for all-cause mortality have been developed previously, there is limited research on factors that may directly affect COPD-specific mortality. Study design-Methods: used probabilistic (causal) graphs to analyze clinical baseline COPDGene data, including demographics, spirometry, quantitative chest imaging, and symptom features, as well as gene expression data (from year-5). Results: We identified factors linked to all-cause and COPD-specific mortality. Although many were similar, there were differences in certain comorbidities (all-cause mortality model only) and forced vital capacity (COPD-specific mortality model only). Using our results, we developed VAPORED , a 7-variable COPD-specific mortality risk score, which we validated using the ECLIPSE 3-yr mortality data. We showed that the new model is more accurate than the existing ADO, BODE, and updated BODE indices. Additionally, we identified biological signatures linked to all-cause mortality, including a plasma cell mediated component. Finally, we developed a web page to help clinicians calculate mortality risk using VAPORED, ADO, and BODE indices. Interpretation: Given the importance of predicting COPD-specific and all-cause mortality risk in COPD patients, we showed that probabilistic graphs can identify the features most directly affecting them, and be used to build new, more accurate models of mortality risk. Novel biological features affecting mortality were also identified. This is an important step towards improving our identification of high-risk patients and potential biological mechanisms that drive COPD mortality.

Joint clinical and molecular subtyping of COPD with variational autoencoders.

Chronic Obstructive Pulmonary Disease (COPD) is a complex, heterogeneous disease. Traditional subtyping methods generally focus on either the clinical manifestations or the molecular endotypes of the disease, resulting in classifications that do not fully capture the disease's complexity. Here, we bridge this gap by introducing a subtyping pipeline that integrates clinical and gene expression data with variational autoencoders. We apply this methodology to the COPDGene study, a large study of current and former smoking individuals with and without COPD. Our approach generates a set of vector embeddings, called Personalized Integrated Profiles (PIPs), that recapitulate the joint clinical and molecular state of the subjects in the study. Prediction experiments show that the PIPs have a predictive accuracy comparable to or better than other embedding approaches. Using trajectory learning approaches, we analyze the main trajectories of variation in the PIP space and identify five well-separated subtypes with distinct clinical phenotypes, expression signatures, and disease outcomes. Notably, these subtypes are more robust to data resampling compared to those identified using traditional clustering approaches. Overall, our findings provide new avenues to establish fine-grained associations between the clinical characteristics, molecular processes, and disease outcomes of COPD.

Machine learning models for blood pressure phenotypes combining multiple polygenic risk scores.

We construct non-linear machine learning (ML) prediction models for systolic and diastolic blood pressure (SBP, DBP) using demographic and clinical variables and polygenic risk scores (PRSs). We developed a two-model ensemble, consisting of a baseline model, where prediction is based on demographic and clinical variables only, and a genetic model, where we also include PRSs. We evaluate the use of a linear versus a non-linear model at both the baseline and the genetic model levels and assess the improvement in performance when incorporating multiple PRSs. We report the ensemble model's performance as percentage variance explained (PVE) on a held-out test dataset. A non-linear baseline model improved the PVEs from 28.1% to 30.1% (SBP) and 14.3% to 17.4% (DBP) compared with a linear baseline model. Including seven PRSs in the genetic model computed based on the largest available GWAS of SBP/DBP improved the genetic model PVE from 4.8% to 5.1% (SBP) and 4.7% to 5% (DBP) compared to using a single PRS. Adding additional 14 PRSs computed based on two independent GWASs further increased the genetic model PVE to 6.3% (SBP) and 5.7% (DBP). PVE differed across self-reported race/ethnicity groups, with primarily all non-White groups benefitting from the inclusion of additional PRSs.