Rajat Deo

Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia and a major cause of cardioembolic stroke. Although polygenic risk scores (PRS) are well characterized to quantify inherited susceptibility for AF, they provide limited insight into the pathways and tissues underlying genetic risk, which are critical to uncover for individual risk prediction. In this study, we develop a pathway-level multi-omics representation learning framework that converts individual genetic profiles into interpretable biological features by integrating GWAS-derived pathway burden scores with tissue-specific transcriptomic pathway signals. We constructed machine learning models to assess population-level AF risk prediction performance across genomic and transcriptomic tissue contexts; the pathway-based global attention models substantially improved risk prediction performance over PRS and other baselines (AUROC improved from 0.601 to 0.738). Transformer and graph neural network frameworks then assessed individual-level pathway interpretability, revealing heterogeneous contributions from electrical signaling, cardiac development, and DNA repair pathways to AF risk. This added interpretability highlights the potential of this pathway approach to enable more mechanistically informed risk stratification than static PRS by capturing underlying heterogeneity. To independently assess whether prioritized pathways reflected cardiac regulatory biology, we compared pathway rankings with transcriptional effects predicted by the AlphaGenome foundation model. Variants in highly ranked pathways showed significantly greater predicted effects on expression in atrial and ventricular tissues (FDR = 0.032) relative to controls, providing orthogonal evidence that the model identifies biologically relevant mechanisms. Overall, this work reframes polygenic risk from a single measure of susceptibility to tissue-informed pathway mechanisms, providing a framework for interpretable genomic stratification in complex diseases.

Was not provided for this article.

Background: Inherited cardiomyopathies cause ventricular tachycardia (VT) and heart failure, yet genotype‐based substrate characterization is limited. Methods: Patients with inherited cardiomyopathy undergoing VT (n=80) or premature ventricular complex (n=25) ablation between 2010 and 2024 were evaluated for arrhythmia substrate, ablation, and heart failure outcomes. Results: Among 105 patients, the median age was 51 years, 72% were men, and the cohort comprised desmosomal (desmoplakin [DSP], 11%; non‐DSP, 24%), titin (22%), lamin A/C (LMNA; 20%), sarcomeric (14%), ion‐channel (5%), and cytoskeletal/Z‐disk (4%) genotype groups. In the VT cohort, low‐voltage substrate was predominantly septal in the LMNA (55%), titin (67%), and sarcomeric (75%) groups; both septal (75%) and lateral left ventricle (75%) in the cytoskeletal/Z‐disk group; perimitral (71%) and lateral left ventricle (57%) in the DSP desmosomal group; and right ventricular free wall (91%) in the non‐DSP desmosomal group. Ablation eliminated clinical VT in 90%, with noninducibility of any VT in 52 of 80 (65%) cases. VT‐free survival was 51% during 3 years (1.8–6.3 years). Among recurrences, 11 of 39 (28%) patients had a single VT episode. VT‐free survival was highest in the non‐DSP desmosomal group and lowest in the LMNA group (66% versus 20%, P =0.03). Residual VT inducibility (hazard ratio [HR], 2 [95% CI, 1.05–3.77]) and LMNA variant (HR, 2.25 [95% CI, 1.1–4.57]) predicted recurrence. In the premature ventricular complex cohort, the burden decreased from 12% (interquartile range [IQR]=Q3–Q1, 21–8) to 3.75% (IQR 3–1, 6–1), and 17 of 25 (68%) had ≤5% burden. End‐stage heart failure outcome (left ventricular assist device, transplant, or death) occurred in 29% overall and 52% in the LMNA group. Conclusions: Genotype correlates with VT substrate. Ablation successfully reduced VT and premature ventricular complex burden; however, long‐term VT recurrence was common except in non‐DSP desmosomal variants. LMNA variants portend worse outcomes.

Background: Coronary artery calcium (CAC) scores inform subclinical atherosclerotic cardiovascular disease (ASCVD) burden, helping guide preventative treatments. However, prediction of cardiovascular (CV) events by CAC is largely limited to ASCVD outcomes. This study investigated whether a previously validated proteomic test for predicting a broad composite of four-year CV events could enhance the prognostic utility of CAC. Methods: We used a 27-protein CV risk score (Prot-CVR), derived from ∼5,000 SomaScan™ Assay plasma protein measurements, to predict four-year risk of a composite CV and mortality outcome (myocardial infarction, stroke/TIA, heart failure hospitalization, death) in 2,122 participants with ≥1 CV risk factors from the Multi-Ethnic Study of Atherosclerosis (MESA) observational cohort at exam 5 and compared predictions to CAC Agatston scores. Discriminatory performance was assessed using C-Index and 4-year area under the curve (AUC). Cox Proportional Hazard (CoxPH) ratios were calculated for the composite outcome, ASCVD outcome (myocardial infarction, resuscitated cardiac arrest, stroke, coronary heart disease death), and individual events. Changes in Prot-CVR and CAC scores from baseline to MESA exam 5 (+10-years) in CV event versus event-free participants were assessed using 2-tailed paired t-tests. CoxPH regression models of CV event status distributed by Prot-CVR, CAC, and relevant co-variates were evaluated for performance relative to individual models. Results: Individual Prot-CVR and CAC models predicting the composite outcome had comparable 4-year AUCs, but Prot-CVR had a higher C-index (0.68 (0.65-0.70) versus 0.63 (0.60-0.65), p=0.001) and greater hazard ratios for the composite outcome (p<0.001), death (p<0.001), and heart failure (p=0.015). A combined CoxPH model of Prot-CVR + CAC + Age had a higher 4-year AUC (0.72, p<0.05) and C-Index (0.71, p<0.05) than Prot-CVR or CAC alone. Both Prot-CVR and CAC scores detected an increase in risk prior to an approaching CV event in ∼10-year sensitivity-to-change analysis. For 49.6% of MESA population with CAC=0 at baseline, Prot-CVR was greater in composite event versus event free participants at 4 years (0.23 versus 0.15, p=0.006) and full follow-up (0.18 versus 0.13, p<0.001). Conclusion: Protein testing complements CAC for CV risk assessment although the improvement is modest. Prot-CVR may resolve which patients with CAC=0 are at heightened CV risk.

Key Points: Using proteomics and Mendelian randomization, we identified causal proteins for the development of atrial fibrillation in patients with CKD. Immunologic function, cell growth, and neural development were common biologic pathways underlying atrial fibrillation and left atrial enlargement. Machine learning led to a nine-protein risk model that had a similar predictive performance for incident atrial fibrillation as clinical risk scores. Background: CKD is strongly associated with atrial fibrillation. Understanding the biologic pathways for this association and creating predictive models have been challenging. Left atrial enlargement is a substrate for atrial fibrillation, but any overlap between biomarkers of atrial fibrillation and left atrial enlargement in patients with CKD is unknown. Methods: We evaluated 4590 plasma proteins with SomaScan in two cohorts of adult patients with CKD: the Chronic Renal Insufficiency Cohort (CRIC; n =2654) and the Atherosclerosis Risk in Communities cohort (ARIC; n =1326). Using Mendelian randomization, we identified proteins along the causal pathway to atrial fibrillation. We also identified proteins and corresponding pathways associated with larger echocardiographic left atrial size, a recognized substrate for atrial fibrillation. Finally, we developed and validated a multiprotein risk score for incident atrial fibrillation in the CKD population. Results: Over 5 years, incident atrial fibrillation occurred among 150 patients in the CRIC and 140 in the ARIC cohort. We identified three proteins causally linked to incident atrial fibrillation: neural EGF like‑like protein 1, cartilage intermediate layer protein 2, and matrix metallopeptidase 12. Pathway analysis revealed an overlap in eight of the top ten canonical pathways for incident atrial fibrillation and left atrial enlargement. A risk model for incident atrial fibrillation composed of proteins had annualized areas under the receiver-operating characteristic curve over 5 years ranging from 0.65 to 0.76, a performance similar to the Cohorts for Heart and Aging Research in Genomic Epidemiology Atrial Fibrillation (CHARGE-AF) clinical risk score. Conclusions: This study identified causal proteins and biologic mechanisms underlying incident atrial fibrillation in CKD. A proteomic risk score for incident atrial fibrillation in CKD performed similarly to the CHARGE-AF clinical risk score.

BACKGROUND: The subcutaneous implantable cardioverter defibrillator (S-ICD) offers protection from sudden cardiac death without transvenous leads. Although contemporary techniques and programming have reduced inappropriate shocks, high rates persist in certain populations. The objective of this study was to evaluate the impact of a novel quantitative vector screening (QVS) protocol on the incidence of sensing-related complications and inappropriate shocks in patients undergoing S-ICD implantation. METHODS: We analyzed 223 consecutive patients who underwent S-ICD implantation at the Hospital of the University of Pennsylvania from December 2018 to July 2025. Traditional vector screening was used before 2023. In 2023, we implemented QVS, which incorporated quantitative sensing scores for each candidate and raised the threshold for S-ICD implantation. The primary end point was time to first inappropriate shock or under-sensed ventricular arrhythmia. Secondary outcomes included SMART Pass deactivation and need for device revision. Outcomes were reported as survival analyses. RESULTS: During preimplant screening, the QVS protocol reduced patient eligibility from 96% to 83%. The median follow-up after implant was 42 months (interquartile range, 48) in the traditional vector screening arm and 18 months (interquartile range, 15) in the QVS arm. The primary end point of time to first inappropriate shock or under-sensed ventricular arrhythmia was longer in the QVS arm (log-rank, P =0.02). There were 23 primary end point events among 145 patients in the traditional vector screening arm (5.2 per 100 patient-years [95% CI, 3.1–7.4]) and 2 primary end point events among 78 patients in the QVS arm (1.8 per 100 patient-years [95% CI, 0.01–4.38]). CONCLUSIONS: Implementation of a novel S-ICD screening protocol with stricter eligibility thresholds reduced sensing-related complications, particularly inappropriate shocks.

Background: This study investigates whether proteomic aging clocks (PACs) are associated with cerebral small vessel disease (CSVD). Methods: We included participants from two US community-based cohorts: the Atherosclerosis Risk in Communities (ARIC) Study and the Multi-Ethnic Study of Atherosclerosis (MESA) Study. These analyses leveraged PACs that were developed in ARIC using proteomics measured by SomaScan in midlife (Visit 2; mean age 56 y; n=1,486) and late-life (Visit 5; mean age 76 y; n=1,496), trained on chronological age. Proteomic age acceleration (PAA) was calculated as residuals from regressing PACs on chronological age. 3T brain MRI data were collected in late-life. We examined associations of PAA with log-transformed white matter hyperintensity (WMH) volume using linear regression and with the presence of microbleeds, and subcortical, lacunar, and cortical infarcts using logistic regression. Associations of PACs with WMH volume and microbleeds were tested in MESA using proteins measured at Exam 1 (mean age 57 y; n=932) and Exam 5 (mean age 66 y; n=934). All associations were quantified per 5-year increase in PAA. All models were adjusted for demographics and cardiovascular risk factors. Results: In ARIC, higher midlife PAA was associated with greater WMH volume (percent difference: 25% [95% CI: 13%, 39%]) and higher odds of subcortical infarcts (OR: 1.24 [1.02, 1.51]). Late-life PAA was associated with all CSVD markers: WMH volume (percent difference: 20% [8%, 34%]), cerebral microbleeds (OR: 1.40 [1.15, 1.69]), subcortical (OR: 1.80 [1.47, 2.22]), lacunar (OR: 1.80 [1.46, 2.23]), and cortical infarcts (OR: 1.39 [1.07, 1.82]). In MESA, higher late-life PAA was associated with greater WMH volume (28% [3%, 58%]) but not with microbleeds. Conclusion: Accelerated proteomic aging is associated with a higher prevalence of MRI markers of CSVD, most predominantly in late-life. Understanding this relationship may help stratify those at higher risk of CSVD at an early stage.