Non-linear machine learning models incorporating SNPs and PRS improve polygenic prediction in diverse human populations

Ge-SAND: an explainable deep learning-driven framework for disease risk prediction by uncovering complex genetic interactions in parallel

BACKGROUND

Accurate genetic risk prediction and understanding the mechanisms underlying complex diseases are essential for effective intervention and precision medicine. However, current methods often struggle to capture the intricate and subtle genetic interactions contributing to disease risk. This challenge may be further exacerbated by the curse of dimensionality when considering large-scale pairwise genetic combinations with limited samples. Overcoming these limitations could transform biomedicine by providing deeper insights into disease mechanisms, moving beyond black-box models and single-locus analyses, and enabling a more comprehensive understanding of cross-disease patterns.

RESULTS

We developed Ge-SAND (Genomic Embedding Self-Attention Neurodynamic Decoder), an explainable deep learning-driven framework designed to uncover complex genetic interactions at scales exceeding 106 in parallel for accurate disease risk prediction. Ge-SAND leverages genotype and genomic positional information to identify both intra- and interchromosomal interactions associated with disease phenotypes, providing comprehensive insights into pathogenic mechanisms crucial for disease risk prediction. Applied to simulated datasets and UK Biobank cohorts for Crohn's disease, schizophrenia, and Alzheimer's disease, Ge-SAND achieved up to a 20% improvement in AUC-ROC compared to mainstream methods. Beyond its predictive accuracy, through self-attention-based interaction networks, Ge-SAND provided insights into large-scale genotype relationships and revealed genetic mechanisms underlying these complex diseases. For instance, Ge-SAND identified potential genetic interaction pairs, including novel relationships such as ISOC1 and HOMER2, potentially implicating the brain-gut axis in Crohn's and Alzheimer's diseases.

CONCLUSION

Ge-SAND is a novel deep-learning approach designed to address the challenges of capturing large-scale genetic interactions. By integrating disease risk prediction with interpretable insights into genetic mechanisms, Ge-SAND offers a valuable tool for advancing genomic research and precision medicine.

Polygenic Risk Scores in Myocardial Injury After Noncardiac Surgery: A VISION Substudy

BACKGROUND

Myocardial injury after noncardiac surgery (MINS) is the most prevalent vascular complication following surgical procedures. Although the revised cardiac risk index (RCRI) is widely used to predict postoperative cardiovascular complications, its predictive accuracy is suboptimal.

OBJECTIVES

Considering genetic influences may improve risk prediction. The authors propose integrating polygenic risk scores (PRS) with the RCRI to enhance MINS prediction. Identification of PRS associated with MINS could provide pathophysiological insights.

METHODS

This is a case-control study nested within the Vascular Events in Noncardiac Surgery Participants Cohort Evaluation cohort, including patients aged 45 and above who underwent noncardiac surgery. Daily troponin levels were measured preoperatively and on days 1, 2, and 3 postoperatively. PRS was computed for MINS risk factors using publicly available summary statistics. Logistic regression models were used to assess the association between each PRS and MINS. PRS discrimination was assessed independently and in combination with RCRI.

RESULTS

A total of 253 MINS cases were matched with 253 controls, adjusted for age, sex, and limited to individuals of European ancestry (ntotal = 506). The type II diabetes (T2D) PRS (OR: 1.26; 95% CI: 1.00-1.58; P = 0.047) and the HbA1c PRS (OR: 1.26; 95% CI: 1.03-1.54; P = 0.026) were associated with MINS. No other PRS, including those for coronary artery disease, stroke, and lipid biomarkers, showed significant associations.

CONCLUSIONS

The T2D PRS and the HbA1c PRS were associated with an increased risk of MINS. The findings may reflect the multifactorial pathophysiology of MINS. Larger genetic studies and trials evaluating perioperative glucose management warrant consideration.

WMH Contributions to Cognitive Impairment: Rationale and Design of the Diverse VCID Study

As awareness of dementia increases, more individuals with minor cognitive complaints are requesting clinical assessment. Neuroimaging studies frequently identify incidental white matter hyperintensities, raising patient concerns about their brain health and future risk for dementia. Moreover, current US demographics indicate that ≈50% of these individuals will be from diverse backgrounds by 2060. Racial and ethnic minority populations bear a disproportionate burden of vascular risk factors magnifying dementia risk. Despite established associations between white matter hyperintensities and cognitive impairment, including dementia, no study has comprehensively and prospectively examined the impact of individual and combined magnetic resonance imaging measures of white matter injury, their risk factors, and comorbidities on cognitive performance among a diverse, nondemented, stroke-free population with cognitive complaints over an extended period of observation. The Diverse VCID (Diverse Vascular Cognitive Impairment and Dementia) study is designed to fill this knowledge gap through 3 assessments of clinical, behavioral, and risk factors; neurocognitive and magnetic resonance imaging measures; fluid biomarkers of Alzheimer disease, vascular inflammation, angiogenesis, and endothelial dysfunction; and measures of genetic risk collected prospectively over a minimum of 3 years in a cohort of 2250 individuals evenly distributed among Americans of Black/African, Latino/Hispanic, and non-Hispanic White backgrounds. The goal of this study is to investigate the basic mechanisms of small vessel cerebrovascular injury, emphasizing clinically relevant assessment tools and developing a risk score that will accurately identify at-risk individuals for possible treatment or clinical therapeutic trials, particularly individuals of diverse backgrounds where vascular risk factors and disease are more prevalent.

Variational Autoencoder-based Model Improves Polygenic Prediction in Blood Cell Traits

Genetic prediction of complex traits, enabled by large-scale genomic studies, has created new measures to understand individual genetic predisposition. Polygenic Risk Scores (PRS) offer a way to aggregate information across the genome, enabling personalized risk prediction for complex traits and diseases. However, conventional PRS calculation methods that rely on linear models are limited in their ability to capture complex patterns and interaction effects in high-dimensional genomic data. In this study, we seek to improve the predictive power of PRS through applying advanced deep learning techniques. We show that the Variational AutoEncoder-based model for PRS construction (VAE-PRS) outperforms currently state-of-the-art methods for biobank-level data in 14 out of 16 blood cell traits, while being computationally efficient. Through comprehensive experiments, we found that the VAE-PRS model offers the ability to capture interaction effects in high-dimensional data and shows robust performance across different pre-screened variant sets. Furthermore, VAE-PRS is easily interpretable via assessing the contribution of each individual marker to the final prediction score through the SHapley Additive exPlanations (SHAP) method, providing potential new insights in identifying trait-associated genetic variants. In summary, VAE-PRS presents a novel measure to genetic risk prediction by harnessing the power of deep learning methods, which could further facilitate the development of personalized medicine and genetic research.

Advancements in genetic research by the Hispanic Community Health Study/Study of Latinos: A 10-year retrospective review

The Hispanic Community Health Study/Study of Latinos (HCHS/SOL) is a multicenter, longitudinal cohort study designed to evaluate environmental, lifestyle, and genetic risk factors as they relate to cardiometabolic and other chronic diseases among Hispanic/Latino populations in the United States. Since the study's inception in 2008, as a result of the study's robust genetic measures, HCHS/SOL has facilitated major contributions to the field of genetic research. This 10-year retrospective review highlights the major findings for genotype-phenotype relationships and advancements in statistical methods owing to the HCHS/SOL. Furthermore, we discuss the ethical and societal challenges of genetic research, especially among Hispanic/Latino adults in the United States. Continued genetic research, ancillary study expansion, and consortia collaboration through HCHS/SOL will further drive knowledge and advancements in human genetics research.

One score to rule them all: regularized ensemble polygenic risk prediction with GWAS summary statistics

Ensemble learning has been increasingly popular for boosting the predictive power of polygenic risk scores (PRS), with almost every recent multi-ancestry PRS approach employing ensemble learning as a final step. Existing ensemble approaches rely on individual-level data for model training, which severely limits their real-world applications, especially in non-European populations without sufficient genomic samples. Here, we introduce a statistical framework to construct regularized ensemble PRS, which allows us to combine a large number of candidate PRS models using only summary statistics from genome-wide association studies. We demonstrate its robust and substantial improvement over many existing PRS models in both within- and cross-ancestry applications. We believe this is truly "one score to rule them all" due to its capability to continuously combine newly developed PRS models with existing models to improve prediction performance, which makes it a universal approach that should always be employed in future PRS applications.

Integrated Germline and Somatic Features Reveal Divergent Immune Pathways Driving Response to Immune Checkpoint Blockade

Immune checkpoint blockade (ICB) has revolutionized cancer treatment; however, the mechanisms determining patient response remain poorly understood. Here, we used machine learning to predict ICB response from germline and somatic biomarkers and interpreted the learned model to uncover putative mechanisms driving superior outcomes. Patients with higher infiltration of T-follicular helper cells had responses even in the presence of defects in the MHC class-I (MHC-I). Further investigation uncovered different ICB responses in tumors when responses were reliant on MHC-I versus MHC-II neoantigens. Despite similar response rates, MHC II-reliant responses were associated with significantly longer durable clinical benefits (discovery: median overall survival of 63.6 vs. 34.5 months; P = 0.0074; validation: median overall survival of 37.5 vs. 33.1 months; P = 0.040). Characteristics of the tumor immune microenvironment reflected MHC neoantigen reliance, and analysis of immune checkpoints revealed LAG3 as a potential target in MHC II-reliant but not MHC I-reliant responses. This study highlights the value of interpretable machine learning models in elucidating the biological basis of therapy responses.

Enhancing Equity in Genomics: Incorporating Measures of Structural Racism, Discrimination, and Social Determinants of Health

The everyday harms of structural racism and discrimination, perpetuated through institutions, laws, policies, and practices, constitute social determinants of health, but measures that account for their debilitating effects are largely missing in genetic studies of complex diseases. Drawing on insights from the social sciences and public health, we propose critical methodologies for incorporating tools that measure structural racism and discrimination within genetic analyses. We illustrate how including these measures may strengthen the accuracy and utility of findings for diverse communities, clarify elusive relationships between genetics and environment in a racialized society, and support greater equity within genomics and precision health research. This approach may also support efforts to build and sustain vital partnerships with communities and with other fields of research inquiry, centering community expertise and lived experiences and drawing on valuable knowledge from practitioners in the social sciences and public health to innovate biomedical and genomic study designs aimed at community health priorities.

Examining the importance of neighborhood natural, and built environment factors in predicting older adults' mental well-being: An XGBoost-SHAP approach

BACKGROUND

Previous studies have shown that urban neighborhood environmental factors significantly influence the health outcomes of urban older adults. However, most cross-sectional studies exploring the health effects of these factors have failed to quantify the relative importance of each factor.

METHODS

We use XGBoost machine learning techniques and SHAPley Additive Interpretation (SHAP) to rank the importance of urban neighborhood environmental factors in shaping the mental health of urban older adults. To address self-selection bias in housing choice, we distinguish older adults living in private housing from those living in public as residents in private housing have more freedom to choose where to live.

RESULTS

The results show that both natural and built environmental factors in urban neighborhoods are important predictors of mental well-being scores. Five natural environmental factors (blue space, perceived greenery quantity, NDVI, street view greenness, aesthetic quality) and three built environmental factors (physical activity facilities quality, physical activity facilities quantity, neighborhood disorder) had considerable predictive power for mental well-being scores in two groups. Among them, blue space, perceived greenery quantity and street view greenness quantity became less important after controlling for self-selection bias, possibly because of the unequal distribution of quantity and quality, and the performance of neighborhood disorder, aesthetic quality and physical activity facilities quality was more sensitive in public housing.

CONCLUSIONS

These results highlight the nuanced and differential effects of neighborhood environmental exposures on mental well-being outcomes, depending on housing preferences. The results of this study can provide support for decision makers in urban planning, landscape design and environmental management in order to improve the mental well-being status of urban older adults.

Introducing GWAStic: a user-friendly, cross-platform solution for genome-wide association studies and genomic prediction

Motivation

Advances in genomics have created an insistent need for accessible tools that simplify complex genetic data analysis, enabling researchers across fields to harness the power of genome-wide association studies and genomic prediction. GWAStic was developed to bridge this gap, providing an intuitive platform that combines artificial intelligence with traditional statistical methods, making sophisticated genomic analysis accessible without requiring deep expertise in statistical software.

Results

We present GWAStic, an intuitive, cross-platform desktop application designed to streamline genome-wide association studies and genomic prediction for biological and medical researchers. With a user-friendly graphical interface, GWAStic integrates machine learning and traditional statistical approaches to support genetic analysis. The application accepts inputs from standard text-based Variant Call Formats and PLINK binary files, generating clear graphical outputs, including Manhattan plots, quantile-quantile plots, and genomic prediction correlation plots to enhance data visualization and analysis.

Availability and implementation

Project page: https://github.com/snowformatics/gwastic_desktop; GWAStic documentation: https://snowformatics.gitbook.io/product-docs; PyPI: https://pypi.org/project/gwastic-desktop/.

Trait imputation enhances nonlinear genetic prediction for some traits

The expansive collection of genetic and phenotypic data within biobanks offers an unprecedented opportunity for biomedical research. However, the frequent occurrence of missing phenotypes presents a significant barrier to fully leveraging this potential. In our target application, on one hand, we have only a small and complete dataset with both genotypes and phenotypes to build a genetic prediction model, commonly called a polygenic (risk) score (PGS or PRS); on the other hand, we have a large dataset of genotypes (e.g. from a biobank) without the phenotype of interest. Our goal is to leverage the large dataset of genotypes (but without the phenotype) and a separate genome-wide association studies summary dataset of the phenotype to impute the phenotypes, which are then used as an individual-level dataset, along with the small complete dataset, to build a nonlinear model as PGS. More specifically, we trained some nonlinear models to 7 imputed and observed phenotypes from the UK Biobank data. We then trained an ensemble model to integrate these models for each trait, resulting in higher R2 values in prediction than using only the small complete (observed) dataset. Additionally, for 2 of the 7 traits, we observed that the nonlinear model trained with the imputed traits had higher R2 than using the imputed traits directly as the PGS, while for the remaining 5 traits, no improvement was found. These findings demonstrate the potential of leveraging existing genetic data and accounting for nonlinear genetic relationships to improve prediction accuracy for some traits.

Improving multiple stressor-response models through the inclusion of nonlinearity and interactions among stressor gradients

Stressor-response models are used to detect and predict changes within ecosystems in response to anthropogenic and naturally occurring stressors. While nonlinear stressor-response relationships and interactions between stressors are common in nature, predictive models often do not account for them due to perceived difficulties in the interpretation of results. We used Irish river monitoring data from 177 river sites to investigate if multiple stressor-response models can be improved by accounting for nonlinearity, interactions in stressor-response relationships and environmental context dependencies. Out of the six models of distinct biological responses, five models benefited from the inclusion of nonlinearity while all six benefited from the inclusion of interactions. The addition of nonlinearity means that we can better see the exponential increase in Trophic Diatom Index (TDI3) as phosphorus increases, inferring ecological conditions deteriorating at a faster rate with increasing phosphorus. Furthermore, our results show that the relationship between stressor and response has the potential to be dependent on other variables, as seen in the interaction of elevation with both siltation and nutrients in relation to Ephemeroptera, Plecoptera and Trichoptera (EPT) richness. Both relationships weakened at higher elevations, perhaps demonstrating that there is a decreased capacity for resilience to stressors at lower elevations due to greater cumulative effects. Understanding interactions such as this is vital to managing ecosystems. Our findings provide empirical support for the need to further develop and employ more complex modelling techniques in environmental assessment and management.

Obstructive sleep apnea mediates genetic risk of Diabetes Mellitus: The Hispanic Community Health Study/Study of Latinos

Objective

We sought to evaluate whether obstructive sleep apnea (OSA), and other sleep disorders, increase genetic risk of developing diabetes mellitus (DM).

Research Design and Methods

Using GWAS summary statistics from the DIAGRAM consortium and Million Veteran Program, we developed multi-ancestry Type 2 Diabetes (T2D) polygenic risk scores (T2D-PRSs) useful in admixed Hispanic/Latino individuals. We estimated the association of the T2D-PRS with cross-sectional and incident DM in the Hispanic Community Health Study/Study of Latinos (HCHS/SOL). We conducted a mediation analysis with T2D-PRSs as an exposure, incident DM as an outcome, and OSA as a mediator. Additionally, we performed Mendelian randomization (MR) analysis to assess the causal relationship between T2D and OSA.

Results

Of 12,342 HCHS/SOL participants, at baseline, 48.4% were normoglycemic, 36.6% were hyperglycemic, and 15% had diabetes, and 50.9% identified as female. Mean age was 41.5, and mean BMI was 29.4. T2D-PRSs was strongly associated with baseline DM and with incident DM. At baseline, a 1 SD increase in the primary T2D-PRS had DM adjusted odds ratio (OR) = 2.67, 95% CI [2.40; 2.97] and a higher incident DM rate (incident rate ratio (IRR) = 2.02, 95% CI [1.75; 2.33]). In a stratified analysis based on OSA severity categories the associations were stronger in individuals with mild OSA compared to those with moderate to severe OSA. Mediation analysis suggested that OSA mediates the T2D-PRS association with DM. In two-sample MR analysis, T2D-PRS had a causal effect on OSA, OR = 1.03, 95% CI [1.01; 1.05], and OSA had a causal effect on T2D, with OR = 2.34, 95% CI [1.59; 3.44].

Conclusions

OSA likely mediates genetic effects on T2D.

Polygenic Risk Score-Based Association Analysis Identifies Genetic Comorbidities Associated with Age-Related Hearing Difficulty in Two Independent Samples

PURPOSE

Age-related hearing loss is the most common form of permanent hearing loss that is associated with various health traits, including Alzheimer's disease, cognitive decline, and depression. The present study aims to identify genetic comorbidities of age-related hearing loss. Past genome-wide association studies identified multiple genomic loci involved in common adult-onset health traits. Polygenic risk scores (PRS) could summarize the polygenic inheritance and quantify the genetic susceptibility of complex traits independent of trait expression. The present study conducted a PRS-based association analysis of age-related hearing difficulty in the UK Biobank sample (N = 425,240), followed by a replication analysis using hearing thresholds (HTs) and distortion-product otoacoustic emissions (DPOAEs) in 242 young adults with self-reported normal hearing. We hypothesized that young adults with genetic comorbidities associated with age-related hearing difficulty would exhibit subclinical decline in HTs and DPOAEs in both ears.

METHODS

A total of 111,243 participants reported age-related hearing difficulty in the UK Biobank sample (> 40 years). The PRS models were derived from the polygenic risk score catalog to obtain 2627 PRS predictors across the health spectrum. HTs (0.25-16 kHz) and DPOAEs (1-16 kHz, L1/L2 = 65/55 dB SPL, F2/F1 = 1.22) were measured on 242 young adults. Saliva-derived DNA samples were subjected to low-pass whole genome sequencing, followed by genome-wide imputation and PRS calculation. The logistic regression analyses were performed to identify PRS predictors of age-related hearing difficulty in the UK Biobank cohort. The linear mixed model analyses were performed to identify PRS predictors of HTs and DPOAEs.

RESULTS

The PRS-based association analysis identified 977 PRS predictors across the health spectrum associated with age-related hearing difficulty. Hearing difficulty and hearing aid use PRS predictors revealed the strongest association with the age-related hearing difficulty phenotype. Youth with a higher genetic predisposition to hearing difficulty revealed a subclinical elevation in HTs and a decline in DPOAEs in both ears. PRS predictors associated with age-related hearing difficulty were enriched for mental health, lifestyle, metabolic, sleep, reproductive, digestive, respiratory, hematopoietic, and immune traits. Fifty PRS predictors belonging to various trait categories were replicated for HTs and DPOAEs in both ears.

CONCLUSION

The study identified genetic comorbidities associated with age-related hearing loss across the health spectrum. Youth with a high genetic predisposition to age-related hearing difficulty and other related complex traits could exhibit sub-clinical decline in HTs and DPOAEs decades before clinically meaningful age-related hearing loss is observed. We posit that effective communication of genetic risk, promoting a healthy lifestyle, and reducing exposure to environmental risk factors at younger ages could help prevent or delay the onset of age-related hearing difficulty at older ages.

Artificial Intelligence for Cardiovascular Care-Part 1: Advances: JACC Review Topic of the Week

Recent artificial intelligence (AI) advancements in cardiovascular care offer potential enhancements in diagnosis, treatment, and outcomes. Innovations to date focus on automating measurements, enhancing image quality, and detecting diseases using novel methods. Applications span wearables, electrocardiograms, echocardiography, angiography, genetics, and more. AI models detect diseases from electrocardiograms at accuracy not previously achieved by technology or human experts, including reduced ejection fraction, valvular heart disease, and other cardiomyopathies. However, AI's unique characteristics necessitate rigorous validation by addressing training methods, real-world efficacy, equity concerns, and long-term reliability. Despite an exponentially growing number of studies in cardiovascular AI, trials showing improvement in outcomes remain lacking. A number are currently underway. Embracing this rapidly evolving technology while setting a high evaluation benchmark will be crucial for cardiology to leverage AI to enhance patient care and the provider experience.

Optimizing clinico-genomic disease prediction across ancestries: a machine learning strategy with Pareto improvement

BACKGROUND

Accurate prediction of an individual's predisposition to diseases is vital for preventive medicine and early intervention. Various statistical and machine learning models have been developed for disease prediction using clinico-genomic data. However, the accuracy of clinico-genomic prediction of diseases may vary significantly across ancestry groups due to their unequal representation in clinical genomic datasets.

METHODS

We introduced a deep transfer learning approach to improve the performance of clinico-genomic prediction models for data-disadvantaged ancestry groups. We conducted machine learning experiments on multi-ancestral genomic datasets of lung cancer, prostate cancer, and Alzheimer's disease, as well as on synthetic datasets with built-in data inequality and distribution shifts across ancestry groups.

RESULTS

Deep transfer learning significantly improved disease prediction accuracy for data-disadvantaged populations in our multi-ancestral machine learning experiments. In contrast, transfer learning based on linear frameworks did not achieve comparable improvements for these data-disadvantaged populations.

CONCLUSIONS

This study shows that deep transfer learning can enhance fairness in multi-ancestral machine learning by improving prediction accuracy for data-disadvantaged populations without compromising prediction accuracy for other populations, thus providing a Pareto improvement towards equitable clinico-genomic prediction of diseases.

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.

A polygenic score method boosted by non-additive models

Dominance heritability in complex traits has received increasing recognition. However, most polygenic score (PGS) approaches do not incorporate non-additive effects. Here, we present GenoBoost, a flexible PGS modeling framework capable of considering both additive and non-additive effects, specifically focusing on genetic dominance. Building on statistical boosting theory, we derive provably optimal GenoBoost scores and provide its efficient implementation for analyzing large-scale cohorts. We benchmark it against seven commonly used PGS methods and demonstrate its competitive predictive performance. GenoBoost is ranked the best for four traits and second-best for three traits among twelve tested disease outcomes in UK Biobank. We reveal that GenoBoost improves prediction for autoimmune diseases by incorporating non-additive effects localized in the MHC locus and, more broadly, works best in less polygenic traits. We further demonstrate that GenoBoost can infer the mode of genetic inheritance without requiring prior knowledge. For example, GenoBoost finds non-zero genetic dominance effects for 602 of 900 selected genetic variants, resulting in 2.5% improvements in predicting psoriasis cases. Lastly, we show that GenoBoost can prioritize genetic loci with genetic dominance not previously reported in the GWAS catalog. Our results highlight the increased accuracy and biological insights from incorporating non-additive effects in PGS models.

Unraveling the Variability of Human Satiation: Implications for Precision Obesity Management

Satiation is the physiologic process that regulates meal size and termination, and it is quantified by the calories consumed to reach satiation. Given its role in energy intake, changes in satiation contribute to obesity's pathogenesis. Our study employed a protocolized approach to study the components of food intake regulation including a standardized breakfast, a gastric emptying study, appetite sensation testing, and a satiation measurement by an ad libitummeal test. These studies revealed that satiation is highly variable among individuals, and while baseline characteristics, anthropometrics, body composition and hormones, contribute to this variability, these factors do not fully account for it. To address this gap, we explored the role of a germline polygenic risk score, which demonstrated a robust association with satiation. Furthermore, we developed a machine-learning-assisted gene risk score to predict satiation and leveraged this prediction to anticipate responses to anti-obesity medications. Our findings underscore the significance of satiation, its inherent variability, and the potential of a genetic risk score to forecast it, ultimately allowing us to predict responses to different anti-obesity interventions.

AI-enhanced integration of genetic and medical imaging data for risk assessment of Type 2 diabetes

Type 2 diabetes (T2D) presents a formidable global health challenge, highlighted by its escalating prevalence, underscoring the critical need for precision health strategies and early detection initiatives. Leveraging artificial intelligence, particularly eXtreme Gradient Boosting (XGBoost), we devise robust risk assessment models for T2D. Drawing upon comprehensive genetic and medical imaging datasets from 68,911 individuals in the Taiwan Biobank, our models integrate Polygenic Risk Scores (PRS), Multi-image Risk Scores (MRS), and demographic variables, such as age, sex, and T2D family history. Here, we show that our model achieves an Area Under the Receiver Operating Curve (AUC) of 0.94, effectively identifying high-risk T2D subgroups. A streamlined model featuring eight key variables also maintains a high AUC of 0.939. This high accuracy for T2D risk assessment promises to catalyze early detection and preventive strategies. Moreover, we introduce an accessible online risk assessment tool for T2D, facilitating broader applicability and dissemination of our findings.

Use of Artificial Intelligence in Improving Outcomes in Heart Disease: A Scientific Statement From the American Heart Association

A major focus of academia, industry, and global governmental agencies is to develop and apply artificial intelligence and other advanced analytical tools to transform health care delivery. The American Heart Association supports the creation of tools and services that would further the science and practice of precision medicine by enabling more precise approaches to cardiovascular and stroke research, prevention, and care of individuals and populations. Nevertheless, several challenges exist, and few artificial intelligence tools have been shown to improve cardiovascular and stroke care sufficiently to be widely adopted. This scientific statement outlines the current state of the art on the use of artificial intelligence algorithms and data science in the diagnosis, classification, and treatment of cardiovascular disease. It also sets out to advance this mission, focusing on how digital tools and, in particular, artificial intelligence may provide clinical and mechanistic insights, address bias in clinical studies, and facilitate education and implementation science to improve cardiovascular and stroke outcomes. Last, a key objective of this scientific statement is to further the field by identifying best practices, gaps, and challenges for interested stakeholders.

Benchmarking multi-ancestry prostate cancer polygenic risk scores in a real-world cohort

Prostate cancer is a heritable disease with ancestry-biased incidence and mortality. Polygenic risk scores (PRSs) offer promising advancements in predicting disease risk, including prostate cancer. While their accuracy continues to improve, research aimed at enhancing their effectiveness within African and Asian populations remains key for equitable use. Recent algorithmic developments for PRS derivation have resulted in improved pan-ancestral risk prediction for several diseases. In this study, we benchmark the predictive power of six widely used PRS derivation algorithms, including four of which adjust for ancestry, against prostate cancer cases and controls from the UK Biobank and All of Us cohorts. We find modest improvement in discriminatory ability when compared with a simple method that prioritizes variants, clumping, and published polygenic risk scores. Our findings underscore the importance of improving upon risk prediction algorithms and the sampling of diverse cohorts.

Comparing feature selection and machine learning approaches for predicting CYP2D6 methylation from genetic variation

Introduction

Pharmacogenetics currently supports clinical decision-making on the basis of a limited number of variants in a few genes and may benefit paediatric prescribing where there is a need for more precise dosing. Integrating genomic information such as methylation into pharmacogenetic models holds the potential to improve their accuracy and consequently prescribing decisions. Cytochrome P450 2D6 (CYP2D6) is a highly polymorphic gene conventionally associated with the metabolism of commonly used drugs and endogenous substrates. We thus sought to predict epigenetic loci from single nucleotide polymorphisms (SNPs) related to CYP2D6 in children from the GUSTO cohort.

Methods

Buffy coat DNA methylation was quantified using the Illumina Infinium Methylation EPIC beadchip. CpG sites associated with CYP2D6 were used as outcome variables in Linear Regression, Elastic Net and XGBoost models. We compared feature selection of SNPs from GWAS mQTLs, GTEx eQTLs and SNPs within 2 MB of the CYP2D6 gene and the impact of adding demographic data. The samples were split into training (75%) sets and test (25%) sets for validation. In Elastic Net model and XGBoost models, optimal hyperparameter search was done using 10-fold cross validation. Root Mean Square Error and R-squared values were obtained to investigate each models' performance. When GWAS was performed to determine SNPs associated with CpG sites, a total of 15 SNPs were identified where several SNPs appeared to influence multiple CpG sites.

Results

Overall, Elastic Net models of genetic features appeared to perform marginally better than heritability estimates and substantially better than Linear Regression and XGBoost models. The addition of nongenetic features appeared to improve performance for some but not all feature sets and probes. The best feature set and Machine Learning (ML) approach differed substantially between CpG sites and a number of top variables were identified for each model.

Discussion

The development of SNP-based prediction models for CYP2D6 CpG methylation in Singaporean children of varying ethnicities in this study has clinical application. With further validation, they may add to the set of tools available to improve precision medicine and pharmacogenetics-based dosing.

Machine Learning Models of Polygenic Risk for Enhanced Prediction of Alzheimer Disease Endophenotypes

Background and Objectives

Alzheimer disease (AD) has a polygenic architecture, for which genome-wide association studies (GWAS) have helped elucidate sequence variants (SVs) influencing susceptibility. Polygenic risk score (PRS) approaches show promise for generating summary measures of inherited risk for clinical AD based on the effects of APOE and other GWAS hits. However, existing PRS approaches, based on traditional regression models, explain only modest variation in AD dementia risk and AD-related endophenotypes. We hypothesized that machine learning (ML) models of polygenic risk (ML-PRS) could outperform standard regression-based PRS methods and therefore have the potential for greater clinical utility.

Methods

We analyzed combined data from the Mayo Clinic Study of Aging (n = 1,791) and the Alzheimer's Disease Neuroimaging Initiative (n = 864). An AD PRS was computed for each participant using the top common SVs obtained from a large AD dementia GWAS. In parallel, ML models were trained using those SV genotypes, with amyloid PET burden as the primary outcome. Secondary outcomes included amyloid PET positivity and clinical diagnosis (cognitively unimpaired vs impaired). We compared performance between ML-PRS and standard PRS across 100 training sessions with different data splits. In each session, data were split into 80% training and 20% testing, and then five-fold cross-validation was used within the training set to ensure the best model was produced for testing. We also applied permutation importance techniques to assess which genetic factors contributed most to outcome prediction.

Results

ML-PRS models outperformed the AD PRS (r2 = 0.28 vs r2 = 0.24 in test set) in explaining variation in amyloid PET burden. Among ML approaches, methods accounting for nonlinear genetic influences were superior to linear methods. ML-PRS models were also more accurate when predicting amyloid PET positivity (area under the curve [AUC] = 0.80 vs AUC = 0.63) and the presence of cognitive impairment (AUC = 0.75 vs AUC = 0.54) compared with the standard PRS.

Discussion

We found that ML-PRS approaches improved upon standard PRS for prediction of AD endophenotypes, partly related to improved accounting for nonlinear effects of genetic susceptibility alleles. Further adaptations of the ML-PRS framework could help to close the gap of remaining unexplained heritability for AD and therefore facilitate more accurate presymptomatic and early-stage risk stratification for clinical decision-making.

Enhancing EEG-based attachment style prediction: unveiling the impact of feature domains

Introduction

Attachment styles are crucial in human relationships and have been explored through neurophysiological responses and EEG data analysis. This study investigates the potential of EEG data in predicting and differentiating secure and insecure attachment styles, contributing to the understanding of the neural basis of interpersonal dynamics.

Methods

We engaged 27 participants in our study, employing an XGBoost classifier to analyze EEG data across various feature domains, including time-domain, complexity-based, and frequency-based attributes.

Results

The study found significant differences in the precision of attachment style prediction: a high precision rate of 96.18% for predicting insecure attachment, and a lower precision of 55.34% for secure attachment. Balanced accuracy metrics indicated an overall model accuracy of approximately 84.14%, taking into account dataset imbalances.

Discussion

These results highlight the challenges in using EEG patterns for attachment style prediction due to the complex nature of attachment insecurities. Individuals with heightened perceived insecurity predominantly aligned with the insecure attachment category, suggesting a link to their increased emotional reactivity and sensitivity to social cues. The study underscores the importance of time-domain features in prediction accuracy, followed by complexity-based features, while noting the lesser impact of frequency-based features. Our findings advance the understanding of the neural correlates of attachment and pave the way for future research, including expanding demographic diversity and integrating multimodal data to refine predictive models.

Integrated germline and somatic features reveal divergent immune pathways driving ICB response

Immune Checkpoint Blockade (ICB) has revolutionized cancer treatment, however mechanisms determining patient response remain poorly understood. Here we used machine learning to predict ICB response from germline and somatic biomarkers and interpreted the learned model to uncover putative mechanisms driving superior outcomes. Patients with higher T follicular helper infiltrates were robust to defects in the class-I Major Histocompatibility Complex (MHC-I). Further investigation uncovered different ICB responses in MHC-I versus MHC-II neoantigen reliant tumors across patients. Despite similar response rates, MHC-II reliant responses were associated with significantly longer durable clinical benefit (Discovery: Median OS=63.6 vs. 34.5 months P=0.0074; Validation: Median OS=37.5 vs. 33.1 months, P=0.040). Characteristics of the tumor immune microenvironment reflected MHC neoantigen reliance, and analysis of immune checkpoints revealed LAG3 as a potential target in MHC-II but not MHC-I reliant responses. This study highlights the value of interpretable machine learning models in elucidating the biological basis of therapy responses.

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.

Artificial Intelligence for Risk Assessment on Primary Prevention of Coronary Artery Disease

Purpose of Review

Coronary artery disease (CAD) is a common and etiologically complex disease worldwide. Current guidelines for primary prevention, or the prevention of a first acute event, include relatively simple risk assessment and leave substantial room for improvement both for risk ascertainment and selection of prevention strategies. Here, we review how advances in big data and predictive modeling foreshadow a promising future of improved risk assessment and precision medicine for CAD.

Recent Findings

Artificial intelligence (AI) has improved the utility of high dimensional data, providing an opportunity to better understand the interplay between numerous CAD risk factors. Beyond applications of AI in cardiac imaging, the vanguard application of AI in healthcare, recent translational research is also revealing a promising path for AI in multi-modal risk prediction using standard biomarkers, genetic and other omics technologies, a variety of biosensors, and unstructured data from electronic health records (EHRs). However, gaps remain in clinical validation of AI models, most notably in the actionability of complex risk prediction for more precise therapeutic interventions.

Summary

The recent availability of nation-scale biobank datasets has provided a tremendous opportunity to richly characterize longitudinal health trajectories using health data collected at home, at laboratories, and through clinic visits. The ever-growing availability of deep genotype-phenotype data is poised to drive a transition from simple risk prediction algorithms to complex, “data-hungry,” AI models in clinical decision-making. While AI models provide the means to incorporate essentially all risk factors into comprehensive risk prediction frameworks, there remains a need to wrap these predictions in interpretable frameworks that map to our understanding of underlying biological mechanisms and associated personalized intervention. This review explores recent advances in the role of machine learning and AI in CAD primary prevention and highlights current strengths as well as limitations mediating potential future applications.

Harmonization of supervised machine learning practices for efficient source attribution of Listeria monocytogenes based on genomic data

BACKGROUND

Genomic data-based machine learning tools are promising for real-time surveillance activities performing source attribution of foodborne bacteria such as Listeria monocytogenes. Given the heterogeneity of machine learning practices, our aim was to identify those influencing the source prediction performance of the usual holdout method combined with the repeated k-fold cross-validation method.

METHODS

A large collection of 1 100 L. monocytogenes genomes with known sources was built according to several genomic metrics to ensure authenticity and completeness of genomic profiles. Based on these genomic profiles (i.e. 7-locus alleles, core alleles, accessory genes, core SNPs and pan kmers), we developed a versatile workflow assessing prediction performance of different combinations of training dataset splitting (i.e. 50, 60, 70, 80 and 90%), data preprocessing (i.e. with or without near-zero variance removal), and learning models (i.e. BLR, ERT, RF, SGB, SVM and XGB). The performance metrics included accuracy, Cohen's kappa, F1-score, area under the curves from receiver operating characteristic curve, precision recall curve or precision recall gain curve, and execution time.

RESULTS

The testing average accuracies from accessory genes and pan kmers were significantly higher than accuracies from core alleles or SNPs. While the accuracies from 70 and 80% of training dataset splitting were not significantly different, those from 80% were significantly higher than the other tested proportions. The near-zero variance removal did not allow to produce results for 7-locus alleles, did not impact significantly the accuracy for core alleles, accessory genes and pan kmers, and decreased significantly accuracy for core SNPs. The SVM and XGB models did not present significant differences in accuracy between each other and reached significantly higher accuracies than BLR, SGB, ERT and RF, in this order of magnitude. However, the SVM model required more computing power than the XGB model, especially for high amount of descriptors such like core SNPs and pan kmers.

CONCLUSIONS

In addition to recommendations about machine learning practices for L. monocytogenes source attribution based on genomic data, the present study also provides a freely available workflow to solve other balanced or unbalanced multiclass phenotypes from binary and categorical genomic profiles of other microorganisms without source code modifications.

A polygenic risk score for Alzheimer's disease constructed using APOE-region variants has stronger association than APOE alleles with mild cognitive impairment in Hispanic/Latino adults in the U.S

INTRODUCTION

Polygenic Risk Scores (PRSs) are summaries of genetic risk alleles for an outcome.

METHODS

We used summary statistics from five GWASs of AD to construct PRSs in 4,189 diverse Hispanics/Latinos (mean age 63 years) from the Study of Latinos-Investigation of Neurocognitive Aging (SOL-INCA). We assessed the PRS associations with MCI in the combined set of people and in diverse subgroups, and when including and excluding the APOE gene region. We also assessed PRS associations with MCI in an independent dataset from the Mass General Brigham Biobank.

RESULTS

A simple sum of 5 PRSs ("PRSsum"), each constructed based on a different AD GWAS, was associated with MCI (OR = 1.28, 95% CI [1.14, 1.41]) in a model adjusted for counts of the APOE-[Formula: see text] and APOE-[Formula: see text] alleles. Associations of single-GWAS PRSs were weaker. When removing SNPs from the APOE region from the PRSs, the association of PRSsum with MCI was weaker (OR = 1.17, 95% CI [1.04,1.31] with adjustment for APOE alleles). In all association analyses, APOE-[Formula: see text] and APOE-[Formula: see text] alleles were not associated with MCI.

DISCUSSION

A sum of AD PRSs is associated with MCI in Hispanic/Latino older adults. Despite no association of APOE-[Formula: see text] and APOE-[Formula: see text] alleles with MCI, the association of the AD PRS with MCI is stronger when including the APOE region. Thus, APOE variants different than the classic APOE alleles may be important predictors of MCI in Hispanic/Latino adults.

Genetic data visualization using literature text-based neural networks: Examples associated with myocardial infarction

Data visualization is critical to unraveling hidden information from complex and high-dimensional data. Interpretable visualization methods are critical, especially in the biology and medical fields, however, there are limited effective visualization methods for large genetic data. Current visualization methods are limited to lower-dimensional data and their performance suffers if there is missing data. In this study, we propose a literature-based visualization method to reduce high-dimensional data without compromising the dynamics of the single nucleotide polymorphisms (SNP) and textual interpretability. Our method is innovative because it is shown to (1) preserves both global and local structures of SNP while reducing the dimension of the data using literature text representations, and (2) enables interpretable visualizations using textual information. For performance evaluations, we examined the proposed approach to classify various classification categories including race, myocardial infarction event age groups, and sex using several machine learning models on the literature-derived SNP data. We used visualization approaches to examine clustering of data as well as quantitative performance metrics for the classification of the risk factors examined above. Our method outperformed all popular dimensionality reduction and visualization methods for both classification and visualization, and it is robust against missing and higher-dimensional data. Moreover, we found it feasible to incorporate both genetic and other risk information obtained from literature with our method.

Deep integrative models for large-scale human genomics

Polygenic risk scores (PRSs) are expected to play a critical role in precision medicine. Currently, PRS predictors are generally based on linear models using summary statistics, and more recently individual-level data. However, these predictors mainly capture additive relationships and are limited in data modalities they can use. We developed a deep learning framework (EIR) for PRS prediction which includes a model, genome-local-net (GLN), specifically designed for large-scale genomics data. The framework supports multi-task learning, automatic integration of other clinical and biochemical data, and model explainability. When applied to individual-level data from the UK Biobank, the GLN model demonstrated a competitive performance compared to established neural network architectures, particularly for certain traits, showcasing its potential in modeling complex genetic relationships. Furthermore, the GLN model outperformed linear PRS methods for Type 1 Diabetes, likely due to modeling non-additive genetic effects and epistasis. This was supported by our identification of widespread non-additive genetic effects and epistasis in the context of T1D. Finally, we constructed PRS models that integrated genotype, blood, urine, and anthropometric data and found that this improved performance for 93% of the 290 diseases and disorders considered. EIR is available at https://github.com/arnor-sigurdsson/EIR.

Explainable multi-task learning improves the parallel estimation of polygenic risk scores for many diseases through shared genetic basis

Many complex diseases share common genetic determinants and are comorbid in a population. We hypothesized that the co-occurrences of diseases and their overlapping genetic etiology can be exploited to simultaneously improve multiple diseases' polygenic risk scores (PRS). This hypothesis was tested using a multi-task learning (MTL) approach based on an explainable neural network architecture. We found that parallel estimations of the PRS for 17 prevalent cancers in a pan-cancer MTL model were generally more accurate than independent estimations for individual cancers in comparable single-task learning (STL) models. Such performance improvement conferred by positive transfer learning was also observed consistently for 60 prevalent non-cancer diseases in a pan-disease MTL model. Interpretation of the MTL models revealed significant genetic correlations between the important sets of single nucleotide polymorphisms used by the neural network for PRS estimation. This suggested a well-connected network of diseases with shared genetic basis.

Evaluation of a genetic risk score computed using human chromosomal-scale length variation to predict breast cancer

INTRODUCTION

The ability to accurately predict whether a woman will develop breast cancer later in her life, should reduce the number of breast cancer deaths. Different predictive models exist for breast cancer based on family history, BRCA status, and SNP analysis. The best of these models has an accuracy (area under the receiver operating characteristic curve, AUC) of about 0.65. We have developed computational methods to characterize a genome by a small set of numbers that represent the length of segments of the chromosomes, called chromosomal-scale length variation (CSLV).

METHODS

We built machine learning models to differentiate between women who had breast cancer and women who did not based on their CSLV characterization. We applied this procedure to two different datasets: the UK Biobank (1534 women with breast cancer and 4391 women who did not) and the Cancer Genome Atlas (TCGA) 874 with breast cancer and 3381 without.

RESULTS

We found a machine learning model that could predict breast cancer with an AUC of 0.836 95% CI (0.830.0.843) in the UK Biobank data. Using a similar approach with the TCGA data, we obtained a model with an AUC of 0.704 95% CI (0.702, 0.706). Variable importance analysis indicated that no single chromosomal region was responsible for significant fraction of the model results.

CONCLUSION

In this retrospective study, chromosomal-scale length variation could effectively predict whether or not a woman enrolled in the UK Biobank study developed breast cancer.

Genomic Innovation in Early Life Cardiovascular Disease Prevention and Treatment

Cardiovascular disease (CVD) is a leading cause of morbidity and mortality globally. Although CVD events do not typically manifest until older adulthood, CVD develops gradually across the life-course, beginning with the elevation of risk factors observed as early as childhood or adolescence and the emergence of subclinical disease that can occur in young adulthood or midlife. Genomic background, which is determined at zygote formation, is among the earliest risk factors for CVD. With major advances in molecular technology, including the emergence of gene-editing techniques, along with deep whole-genome sequencing and high-throughput array-based genotyping, scientists now have the opportunity to not only discover genomic mechanisms underlying CVD but use this knowledge for the life-course prevention and treatment of these conditions. The current review focuses on innovations in the field of genomics and their applications to monogenic and polygenic CVD prevention and treatment. With respect to monogenic CVD, we discuss how the emergence of whole-genome sequencing technology has accelerated the discovery of disease-causing variants, allowing comprehensive screening and early, aggressive CVD mitigation strategies in patients and their families. We further describe advances in gene editing technology, which might soon make possible cures for CVD conditions once thought untreatable. In relation to polygenic CVD, we focus on recent innovations that leverage findings of genome-wide association studies to identify druggable gene targets and develop predictive genomic models of disease, which are already facilitating breakthroughs in the life-course treatment and prevention of CVD. Gaps in current research and future directions of genomics studies are also discussed. In aggregate, we hope to underline the value of leveraging genomics and broader multiomics information for characterizing CVD conditions, work which promises to expand precision approaches for the life-course prevention and treatment of CVD.

Germline modifiers of the tumor immune microenvironment implicate drivers of cancer risk and immunotherapy response

With the continued promise of immunotherapy for treating cancer, understanding how host genetics contributes to the tumor immune microenvironment (TIME) is essential to tailoring cancer screening and treatment strategies. Here, we study 1084 eQTLs affecting the TIME found through analysis of The Cancer Genome Atlas and literature curation. These TIME eQTLs are enriched in areas of active transcription, and associate with gene expression in specific immune cell subsets, such as macrophages and dendritic cells. Polygenic score models built with TIME eQTLs reproducibly stratify cancer risk, survival and immune checkpoint blockade (ICB) response across independent cohorts. To assess whether an eQTL-informed approach could reveal potential cancer immunotherapy targets, we inhibit CTSS, a gene implicated by cancer risk and ICB response-associated polygenic models; CTSS inhibition results in slowed tumor growth and extended survival in vivo. These results validate the potential of integrating germline variation and TIME characteristics for uncovering potential targets for immunotherapy.

Deep Learning Framework for Complex Disease Risk Prediction Using Genomic Variations

Genome-wide association studies have proven their ability to improve human health outcomes by identifying genotypes associated with phenotypes. Various works have attempted to predict the risk of diseases for individuals based on genotype data. This prediction can either be considered as an analysis model that can lead to a better understanding of gene functions that underlie human disease or as a black box in order to be used in decision support systems and in early disease detection. Deep learning techniques have gained more popularity recently. In this work, we propose a deep-learning framework for disease risk prediction. The proposed framework employs a multilayer perceptron (MLP) in order to predict individuals' disease status. The proposed framework was applied to the Wellcome Trust Case-Control Consortium (WTCCC), the UK National Blood Service (NBS) Control Group, and the 1958 British Birth Cohort (58C) datasets. The performance comparison of the proposed framework showed that the proposed approach outperformed the other methods in predicting disease risk, achieving an area under the curve (AUC) up to 0.94.

Inferring feature importance with uncertainties with application to large genotype data

Estimating feature importance, which is the contribution of a prediction or several predictions due to a feature, is an essential aspect of explaining data-based models. Besides explaining the model itself, an equally relevant question is which features are important in the underlying data generating process. We present a Shapley-value-based framework for inferring the importance of individual features, including uncertainty in the estimator. We build upon the recently published model-agnostic feature importance score of SAGE (Shapley additive global importance) and introduce Sub-SAGE. For tree-based models, it has the advantage that it can be estimated without computationally expensive resampling. We argue that for all model types the uncertainties in our Sub-SAGE estimator can be estimated using bootstrapping and demonstrate the approach for tree ensemble methods. The framework is exemplified on synthetic data as well as large genotype data for predicting feature importance with respect to obesity.

Ethical layering in AI-driven polygenic risk scores-New complexities, new challenges

Researchers aim to develop polygenic risk scores as a tool to prevent and more effectively treat serious diseases, disorders and conditions such as breast cancer, type 2 diabetes mellitus and coronary heart disease. Recently, machine learning techniques, in particular deep neural networks, have been increasingly developed to create polygenic risk scores using electronic health records as well as genomic and other health data. While the use of artificial intelligence for polygenic risk scores may enable greater accuracy, performance and prediction, it also presents a range of increasingly complex ethical challenges. The ethical and social issues of many polygenic risk score applications in medicine have been widely discussed. However, in the literature and in practice, the ethical implications of their confluence with the use of artificial intelligence have not yet been sufficiently considered. Based on a comprehensive review of the existing literature, we argue that this stands in need of urgent consideration for research and subsequent translation into the clinical setting. Considering the many ethical layers involved, we will first give a brief overview of the development of artificial intelligence-driven polygenic risk scores, associated ethical and social implications, challenges in artificial intelligence ethics, and finally, explore potential complexities of polygenic risk scores driven by artificial intelligence. We point out emerging complexity regarding fairness, challenges in building trust, explaining and understanding artificial intelligence and polygenic risk scores as well as regulatory uncertainties and further challenges. We strongly advocate taking a proactive approach to embedding ethics in research and implementation processes for polygenic risk scores driven by artificial intelligence.

Genetic determinants of cardiometabolic and pulmonary phenotypes and obstructive sleep apnoea in HCHS/SOL

BACKGROUND

Obstructive Sleep Apnoea (OSA) often co-occurs with cardiometabolic and pulmonary diseases. This study is to apply genetic analysis methods to explain the associations between OSA and related phenotypes.

METHODS

In the Hispanic Community Healthy Study/Study of Latinos, we estimated genetic correlations ρ_g between the respiratory event index (REI) and 54 anthropometric, glycemic, cardiometabolic, and pulmonary phenotypes. We used summary statistics from published genome-wide association studies to construct Polygenic Risk Scores (PRSs) representing the genetic basis of each correlated phenotype (ρ_g>0.2 and p-value<0.05), and of OSA. We studied the association of the PRSs of the correlated phenotypes with both REI and OSA (REI≥5), and the association of OSA PRS with the correlated phenotypes. Causal relationships were tested using Mendelian Randomization (MR) analysis.

FINDINGS

The dataset included 11,155 participants, 31.03% with OSA. 22 phenotypes were genetically correlated with REI. 10 PRSs covering obesity and fat distribution (BMI, WHR, WHRadjBMI), blood pressure (DBP, PP, MAP), glycaemic control (fasting insulin, HbA1c, HOMA-B) and insomnia were associated with REI and/or OSA. OSA PRS was associated with BMI, WHR, DBP and glycaemic traits (fasting insulin, HbA1c, HOMA-B and HOMA-IR). MR analysis identified robust causal effects of BMI and WHR on OSA, and probable causal effects of DBP, PP, and HbA1c on OSA/REI.

INTERPRETATION

There are shared genetic underpinnings of anthropometric, blood pressure, and glycaemic phenotypes with OSA, with evidence for causal relationships between some phenotypes.

FUNDING

Described in Acknowledgments.

Non-linear machine learning models incorporating SNPs and PRS improve polygenic prediction in diverse human populations

Polygenic risk scores (PRS) are commonly used to quantify the inherited susceptibility for a trait, yet they fail to account for non-linear and interaction effects between single nucleotide polymorphisms (SNPs). We address this via a machine learning approach, validated in nine complex phenotypes in a multi-ancestry population. We use an ensemble method of SNP selection followed by gradient boosted trees (XGBoost) to allow for non-linearities and interaction effects. We compare our results to the standard, linear PRS model developed using PRSice, LDpred2, and lassosum2. Combining a PRS as a feature in an XGBoost model results in a relative increase in the percentage variance explained compared to the standard linear PRS model by 22% for height, 27% for HDL cholesterol, 43% for body mass index, 50% for sleep duration, 58% for systolic blood pressure, 64% for total cholesterol, 66% for triglycerides, 77% for LDL cholesterol, and 100% for diastolic blood pressure. Multi-ancestry trained models perform similarly to specific racial/ethnic group trained models and are consistently superior to the standard linear PRS models. This work demonstrates an effective method to account for non-linearities and interaction effects in genetics-based prediction models.