Search

Abstract Background The additive model of inheritance assumes that heterozygotes (Aa) are exactly intermediate in respect to homozygotes (AA and aa). While this model is commonly used in single-locus genetic association studies, significant deviations from additivity are well-documented and contribute to phenotypic variance across many traits and systems. This assumption can introduce type I and type II errors by overestimating or underestimating the effects of variants that deviate from additivity. Alternative genotype encoding strategies have been explored to account for different inheritance patterns, but they often incur significant computational or methodological costs. To address these challenges, we introduce PAGER (Phenotype Adjusted Genotype Encoding and Ranking), an efficient pre-processing method that encodes each genetic variant based on normalized mean phenotypic differences between diallelic genotype classes (AA, Aa, and aa). This approach more accurately reflects each variant’s true inheritance model, improving model precision while minimizing the costs associated with alternative encoding strategies. Results Through extensive benchmarking on SNPs simulated with both binary and continuous phenotypes, we demonstrate that PAGER accurately represents various inheritance patterns (including additive, dominant, recessive, and heterosis), achieves levels of statistical power that meet or exceed other encoding strategies, and attains computation speeds up to 55 times faster than a similar method, EDGE. We also apply PAGER to publicly available real-world data and identify a novel, relevant putative QTL associated with body mass index in rats (Rattus norvegicus) that is not detected with the additive model. Conclusions Overall, we show that PAGER is an efficient genotype encoding approach that can uncover sources of missing heritability and reveal novel insights in the study of complex traits while incurring minimal costs.

Abstract Background Epistasis, the interaction between genetic loci where the effect of one locus is influenced by one or more other loci, plays a crucial role in the genetic architecture of complex traits. However, as the number of loci considered increases, the investigation of epistasis becomes exponentially more complex, making the selection of key features vital for effective downstream analyses. Relief-Based Algorithms (RBAs) are often employed for this purpose due to their reputation as “interaction-sensitive” algorithms and uniquely non-exhaustive approach. However, the limitations of RBAs in detecting interactions, particularly those involving multiple loci, have not been thoroughly defined. This study seeks to address this gap by evaluating the efficiency of RBAs in detecting higher-order epistatic interactions. Motivated by previous findings that suggest some RBAs may rank predictive features involved in higher-order epistasis negatively, we explore the potential of absolute value ranking of RBA feature weights as an alternative approach for capturing complex interactions. In this study, we assess the performance of ReliefF, MultiSURF, and MultiSURFstar on simulated genetic datasets that model various patterns of genotype-phenotype associations, including 2-way to 5-way genetic interactions, and compare their performance to two control methods: a random shuffle and mutual information. Results Our findings indicate that while RBAs effectively identify lower-order (2 to 3-way) interactions, their capability to detect higher-order interactions is significantly limited, primarily by large feature count but also by signal noise. Specifically, we observe that RBAs are successful in detecting fully penetrant 4-way XOR interactions using an absolute value ranking approach, but this is restricted to datasets with only 20 total features. Conclusions These results highlight the inherent limitations of current RBAs and underscore the need for the development of Relief-based approaches with enhanced detection capabilities for the investigation of epistasis, particularly in datasets with large feature counts and complex higher-order interactions.

Abstract GPT-4, as the most advanced version of OpenAI’s large language models, has attracted widespread attention, rapidly becoming an indispensable AI tool across various areas. This includes its exploration by scientists for diverse applications. Our study focused on assessing GPT-4’s capabilities in generating text, tables, and diagrams for biomedical review papers. We also assessed the consistency in text generation by GPT-4, along with potential plagiarism issues when employing this model for the composition of scientific review papers. Based on the results, we suggest the development of enhanced functionalities in ChatGPT, aiming to meet the needs of the scientific community more effectively. This includes enhancements in uploaded document processing for reference materials, a deeper grasp of intricate biomedical concepts, more precise and efficient information distillation for table generation, and a further refined model specifically tailored for scientific diagram creation.

Abstract Determining the fundamental characteristics that define a face as "feminine" or "masculine" has long fascinated anatomists and plastic surgeons, particularly those involved in aesthetic and gender-affirming surgery. Previous studies in this area have relied on manual measurements, comparative anatomy, and heuristic landmark-based feature extraction. In this study, we collected retrospectively at Cedars Sinai Medical Center (CSMC) a dataset of 98 skull samples, which is the first dataset of this kind of 3D medical imaging. We then evaluated the accuracy of multiple deep learning neural network architectures on sex classification with this dataset. Specifically, we evaluated methods representing three different 3D data modeling approaches: Resnet3D, PointNet++, and MeshNet. Despite the limited number of imaging samples, our testing results show that all three approaches achieve AUC scores above 0.9 after convergence. PointNet++ exhibits the highest accuracy, while MeshNet has the lowest. Our findings suggest that accuracy is not solely dependent on the sparsity of data representation but also on the architecture design, with MeshNet's lower accuracy likely due to the lack of a hierarchical structure for progressive data abstraction. Furthermore, we studied a problem related to sex determination, which is the analysis of the various morphological features that affect sex classification. We proposed and developed a new method based on morphological gradients to visualize features that influence model decision making. The method based on morphological gradients is an alternative to the standard saliency map, and the new method provides better visualization of feature importance. Our study is the first to develop and evaluate deep learning models for analyzing 3D facial skull images to identify imaging feature differences between individuals assigned male or female at birth. These findings may be useful for planning and evaluating craniofacial surgery, particularly gender-affirming procedures, such as facial feminization surgery.

Abstract Purpose Epistasis, the interaction between two or more genes, is integral to the study of genetics and is present throughout nature. Yet, it is seldom fully explored as most approaches primarily focus on single-locus effects, partly because analyzing all pairwise and higher-order interactions requires significant computational resources. Furthermore, existing methods for epistasis detection only consider a Cartesian (multiplicative) model for interaction terms. This is likely limiting as epistatic interactions can evolve to produce varied relationships between genetic loci, some complex and not linearly separable. Methods We present new algorithms for the interaction coefficients for standard regression models for epistasis that permit many varied models for the interaction terms for loci and efficient memory usage. The algorithms are given for two-way and three-way epistasis and may be generalized to higher order epistasis. Statistical tests for the interaction coefficients are also provided. We also present an efficient matrix based algorithm for permutation testing for two-way epistasis. We offer a proof and experimental evidence that methods that look for epistasis only at loci that have main effects may not be justified. Given the computational efficiency of the algorithm, we applied the method to a rat data set and mouse data set, with at least 10,000 loci and 1,000 samples each, using the standard Cartesian model and the XOR model to explore body mass index. Results This study reveals that although many of the loci found to exhibit significant statistical epistasis overlap between models in rats, the pairs are mostly distinct. Further, the XOR model found greater evidence for statistical epistasis in many more pairs of loci in both data sets with almost all significant epistasis in mice identified using XOR. In the rat data set, loci involved in epistasis under the XOR model are enriched for biologically relevant pathways. Conclusion Our results in both species show that many biologically relevant epistatic relationships would have been undetected if only one interaction model was applied, providing evidence that varied interaction models should be implemented to explore epistatic interactions that occur in living systems.

Abstract In response to the escalating global obesity crisis and its associated health and financial burdens, this paper presents a novel methodology for analyzing longitudinal weight loss data and assessing the effectiveness of financial incentives. Drawing from the Keep It Off trial—a three-arm randomized controlled study with 189 participants—we examined the potential impact of financial incentives on weight loss maintenance. Given that some participants choose not to weigh themselves because of small weight change or weight gains, which is a common phenomenon in many weight-loss studies, traditional methods, for example, the Generalized Estimating Equations (GEE) method tends to overestimate the effect size due to the assumption that data are missing completely at random. To address this challenge, we proposed a framework which can identify evidence of missing not at random and conduct bias correction using the estimating equation derived from pairwise composite likelihood. By analyzing the Keep It Off data, we found that the data in this trial are most likely characterized by non-random missingness. Notably, we also found that the enrollment time (i.e., duration time) would be positively associated with the weight loss maintenance after adjusting for the baseline participant characteristics (e.g., age, sex). Moreover, the lottery-based intervention was found to be more effective in weight loss maintenance compared with the direct payment intervention, though the difference was non-statistically significant. This framework's significance extends beyond weight loss research, offering a semi-parametric approach to assess missing data mechanisms and robustly explore associations between exposures (e.g., financial incentives) and key outcomes (e.g., weight loss maintenance). In essence, the proposed methodology provides a powerful toolkit for analyzing real-world longitudinal data, particularly in scenarios with data missing not at random, enriching comprehension of intricate dataset dynamics.

Abstract There are not currently any univariate outlier detection algorithms that transform and model arbitrarily shaped distributions to remove univariate outliers. Some algorithms model skew, even fewer model kurtosis, and none of them model bimodality and monotonicity. To overcome these challenges, we have implemented an algorithm for Skew and Tail-heaviness Adjusted Removal of Outliers (STAR_outliers) that robustly removes univariate outliers from distributions with many different shape profiles, including extreme skew, extreme kurtosis, bimodality, and monotonicity. We show that STAR_outliers removes simulated outliers with greater recall and precision than several general algorithms, and it also models the outlier bounds of real data distributions with greater accuracy. Background Reliably removing univariate outliers from arbitrarily shaped distributions is a difficult task. Incorrectly assuming unimodality or overestimating tail heaviness fails to remove outliers, while underestimating tail heaviness incorrectly removes regular data from the tails. Skew often produces one heavy tail and one light tail, and we show that several sophisticated outlier removal algorithms often fail to remove outliers from the light tail. Multivariate outlier detection algorithms have recently become popular, but having tested PyOD’s multivariate outlier removal algorithms, we found them to be inadequate for univariate outlier removal. They usually do not allow for univariate input, and they do not fit their distributions of outliership scores with a model on which an outlier threshold can be accurately established. Thus, there is a need for a flexible outlier removal algorithm that can model arbitrarily shaped univariate distributions. Results In order to effectively model arbitrarily shaped univariate distributions, we have combined several well-established algorithms into a new algorithm called STAR_outliers. STAR_outliers removes more simulated true outliers and fewer non-outliers than several other univariate algorithms. These include several normality-assuming outlier removal methods, PyOD’s isolation forest (IF) outlier removal algorithm (ACM Transactions on Knowledge Discovery from Data (TKDD) 6:3, 2012) with default settings, and an IQR based algorithm by Verardi and Vermandele that removes outliers while accounting for skew and kurtosis (Verardi and Vermandele, Journal de la Société Française de Statistique 157:90–114, 2016). Since the IF algorithm’s default model poorly fit the outliership scores, we also compared the isolation forest algorithm with a model that entails removing as many datapoints as STAR_outliers does in order of decreasing outliership scores. We also compared these algorithms on the publicly available 2018 National Health and Nutrition Examination Survey (NHANES) data by setting the outlier threshold to keep values falling within the main 99.3 percent of the fitted model’s domain. We show that our STAR_outliers algorithm removes significantly closer to 0.7 percent of values from these features than other outlier removal methods on average. Conclusions STAR_outliers is an easily implemented python package for removing outliers that outperforms multiple commonly used methods of univariate outlier removal.

Abstract In computational toxicology, prediction of complex endpoints has always been challenging, as they often involve multiple distinct mechanisms. State‐of‐the‐art models are either limited by low accuracy, or lack of interpretability due to their black‐box nature. Here, we introduce AIDTox, an interpretable deep learning model which incorporates curated knowledge of chemical‐gene connections, gene‐pathway annotations, and pathway hierarchy. AIDTox accurately predicts cytotoxicity outcomes in HepG2 and HEK293 cells. It also provides comprehensive explanations of cytotoxicity covering multiple aspects of drug activity, including target interaction, metabolism, and elimination. In summary, AIDTox provides a computational framework for unveiling cellular mechanisms for complex toxicity endpoints.

Abstract The introduction of large language models (LLMs) that allow iterative “chat” in late 2022 is a paradigm shift that enables generation of text often indistinguishable from that written by humans. LLM-based chatbots have immense potential to improve academic work efficiency, but the ethical implications of their fair use and inherent bias must be considered. In this editorial, we discuss this technology from the academic’s perspective with regard to its limitations and utility for academic writing, education, and programming. We end with our stance with regard to using LLMs and chatbots in academia, which is summarized as (1) we must find ways to effectively use them, (2) their use does not constitute plagiarism (although they may produce plagiarized text), (3) we must quantify their bias, (4) users must be cautious of their poor accuracy, and (5) the future is bright for their application to research and as an academic tool.

Abstract Machine learning (ML) models trained for triggering clinical decision support (CDS) are typically either accurate or interpretable but not both. Scaling CDS to the panoply of clinical use cases while mitigating risks to patients will require many ML models be intuitively interpretable for clinicians. To this end, we adapted a symbolic regression method, coined the feature engineering automation tool (FEAT), to train concise and accurate models from high-dimensional electronic health record (EHR) data. We first present an in-depth application of FEAT to classify hypertension, hypertension with unexplained hypokalemia, and apparent treatment-resistant hypertension (aTRH) using EHR data for 1200 subjects receiving longitudinal care in a large healthcare system. FEAT models trained to predict phenotypes adjudicated by chart review had equivalent or higher discriminative performance (p

Abstract Background Quantitative Trait Locus (QTL) analysis and Genome-Wide Association Studies (GWAS) have the power to identify variants that capture significant levels of phenotypic variance in complex traits. However, effort and time are required to select the best methods and optimize parameters and pre-processing steps. Although machine learning approaches have been shown to greatly assist in optimization and data processing, applying them to QTL analysis and GWAS is challenging due to the complexity of large, heterogenous datasets. Here, we describe proof-of-concept for an automated machine learning approach, AutoQTL, with the ability to automate many complicated decisions related to analysis of complex traits and generate solutions to describe relationships that exist in genetic data. Results Using a publicly available dataset of 18 putative QTL from a large-scale GWAS of body mass index in the laboratory rat, Rattus norvegicus, AutoQTL captures the phenotypic variance explained under a standard additive model. AutoQTL also detects evidence of non-additive effects including deviations from additivity and 2-way epistatic interactions in simulated data via multiple optimal solutions. Additionally, feature importance metrics provide different insights into the inheritance models and predictive power of multiple GWAS-derived putative QTL. Conclusions This proof-of-concept illustrates that automated machine learning techniques can complement standard approaches and have the potential to detect both additive and non-additive effects via various optimal solutions and feature importance metrics. In the future, we aim to expand AutoQTL to accommodate omics-level datasets with intelligent feature selection and feature engineering strategies.

Summary: Background: Multisystem inflammatory syndrome in children (MIS-C) is a severe complication of SARS-CoV-2 infection. It remains unclear how MIS-C phenotypes vary across SARS-CoV-2 variants. We aimed to investigate clinical characteristics and outcomes of MIS-C across SARS-CoV-2 eras. Methods: We performed a multicentre observational retrospective study including seven paediatric hospitals in four countries (France, Spain, U.K., and U.S.). All consecutive confirmed patients with MIS-C hospitalised between February 1st, 2020, and May 31st, 2022, were included. Electronic Health Records (EHR) data were used to calculate pooled risk differences (RD) and effect sizes (ES) at site level, using Alpha as reference. Meta-analysis was used to pool data across sites. Findings: Of 598 patients with MIS-C (61% male, 39% female; mean age 9.7 years [SD 4.5]), 383 (64%) were admitted in the Alpha era, 111 (19%) in the Delta era, and 104 (17%) in the Omicron era. Compared with patients admitted in the Alpha era, those admitted in the Delta era were younger (ES −1.18 years [95% CI −2.05, −0.32]), had fewer respiratory symptoms (RD −0.15 [95% CI −0.33, −0.04]), less frequent non-cardiogenic shock or systemic inflammatory response syndrome (SIRS) (RD −0.35 [95% CI −0.64, −0.07]), lower lymphocyte count (ES −0.16 × 109/uL [95% CI −0.30, −0.01]), lower C-reactive protein (ES −28.5 mg/L [95% CI −46.3, −10.7]), and lower troponin (ES −0.14 ng/mL [95% CI −0.26, −0.03]). Patients admitted in the Omicron versus Alpha eras were younger (ES −1.6 years [95% CI −2.5, −0.8]), had less frequent SIRS (RD −0.18 [95% CI −0.30, −0.05]), lower lymphocyte count (ES −0.39 × 109/uL [95% CI −0.52, −0.25]), lower troponin (ES −0.16 ng/mL [95% CI −0.30, −0.01]) and less frequently received anticoagulation therapy (RD −0.19 [95% CI −0.37, −0.04]). Length of hospitalization was shorter in the Delta versus Alpha eras (−1.3 days [95% CI −2.3, −0.4]). Interpretation: Our study suggested that MIS-C clinical phenotypes varied across SARS-CoV-2 eras, with patients in Delta and Omicron eras being younger and less sick. EHR data can be effectively leveraged to identify rare complications of pandemic diseases and their variation over time. Funding: None.

Summary: Background: Characterizing Post-Acute Sequelae of COVID (SARS-CoV-2 Infection), or PASC has been challenging due to the multitude of sub-phenotypes, temporal attributes, and definitions. Scalable characterization of PASC sub-phenotypes can enhance screening capacities, disease management, and treatment planning. Methods: We conducted a retrospective multi-centre observational cohort study, leveraging longitudinal electronic health record (EHR) data of 30,422 patients from three healthcare systems in the Consortium for the Clinical Characterization of COVID-19 by EHR (4CE). From the total cohort, we applied a deductive approach on 12,424 individuals with follow-up data and developed a distributed representation learning process for providing augmented definitions for PASC sub-phenotypes. Findings: Our framework characterized seven PASC sub-phenotypes. We estimated that on average 15.7% of the hospitalized COVID-19 patients were likely to suffer from at least one PASC symptom and almost 5.98%, on average, had multiple symptoms. Joint pain and dyspnea had the highest prevalence, with an average prevalence of 5.45% and 4.53%, respectively. Interpretation: We provided a scalable framework to every participating healthcare system for estimating PASC sub-phenotypes prevalence and temporal attributes, thus developing a unified model that characterizes augmented sub-phenotypes across the different systems. Funding: Authors are supported by National Institute of Allergy and Infectious Diseases, National Institute on Aging, National Center for Advancing Translational Sciences, National Medical Research Council, National Institute of Neurological Disorders and Stroke, European Union, National Institutes of Health, National Center for Advancing Translational Sciences.

Abstract Background Loss-of-Function (LoF) variants in human genes are important due to their impact on clinical phenotypes and frequent occurrence in the genomes of healthy individuals. The association of LoF variants with complex diseases and traits may lead to the discovery and validation of novel therapeutic targets. Current approaches predict high-confidence LoF variants without identifying the specific genes or the number of copies they affect. Moreover, there is a lack of methods for detecting knockout genes caused by compound heterozygous (CH) LoF variants. Results We have developed the Loss-of-Function ToolKit (LoFTK), which allows efficient and automated prediction of LoF variants from genotyped, imputed and sequenced genomes. LoFTK enables the identification of genes that are inactive in one or two copies and provides summary statistics for downstream analyses. LoFTK can identify CH LoF variants, which result in LoF genes with two copies lost. Using data from parents and offspring we show that 96% of CH LoF genes predicted by LoFTK in the offspring have the respective alleles donated by each parent. Conclusions LoFTK is a command-line based tool that provides a reliable computational workflow for predicting LoF variants from genotyped and sequenced genomes, identifying genes that are inactive in 1 or 2 copies. LoFTK is an open software and is freely available to non-commercial users at https://github.com/CirculatoryHealth/LoFTK .

Abstract Background Alzheimer’s disease (AD) is a complex neurodegenerative disorder and the most common type of dementia. AD is characterized by a decline of cognitive function and brain atrophy, and is highly heritable with estimated heritability ranging from 60 to 80 $$\%$$ % . The most straightforward and widely used strategy to identify AD genetic basis is to perform genome-wide association study (GWAS) of the case-control diagnostic status. These GWAS studies have identified over 50 AD related susceptibility loci. Recently, imaging genetics has emerged as a new field where brain imaging measures are studied as quantitative traits to detect genetic factors. Given that many imaging genetics studies did not involve the diagnostic outcome in the analysis, the identified imaging or genetic markers may not be related or specific to the disease outcome. Results We propose a novel method to identify disease-related genetic variants enriched by imaging endophenotypes, which are the imaging traits associated with both genetic factors and disease status. Our analysis consists of three steps: (1) map the effects of a genetic variant (e.g., single nucleotide polymorphism or SNP) onto imaging traits across the brain using a linear regression model, (2) map the effects of a diagnosis phenotype onto imaging traits across the brain using a linear regression model, and (3) detect SNP-diagnosis association via correlating the SNP effects with the diagnostic effects on the brain-wide imaging traits. We demonstrate the promise of our approach by applying it to the Alzheimer’s Disease Neuroimaging Initiative database. Among 54 AD related susceptibility loci reported in prior large-scale AD GWAS, our approach identifies 41 of those from a much smaller study cohort while the standard association approaches identify only two of those. Clearly, the proposed imaging endophenotype enriched approach can reveal promising AD genetic variants undetectable using the traditional method. Conclusion We have proposed a novel method to identify AD genetic variants enriched by brain-wide imaging endophenotypes. This approach can not only boost detection power, but also reveal interesting biological pathways from genetic determinants to intermediate brain traits and to phenotypic AD outcomes.

Abstract Objectives Ascertain and compare the performances of Automated Machine Learning (AutoML) tools on large, highly imbalanced healthcare datasets. Materials and Methods We generated a large dataset using historical de-identified administrative claims including demographic information and flags for disease codes in four different time windows prior to 2019. We then trained three AutoML tools on this dataset to predict six different disease outcomes in 2019 and evaluated model performances on several metrics. Results The AutoML tools showed improvement from the baseline random forest model but did not differ significantly from each other. All models recorded low area under the precision-recall curve and failed to predict true positives while keeping the true negative rate high. Model performance was not directly related to prevalence. We provide a specific use-case to illustrate how to select a threshold that gives the best balance between true and false positive rates, as this is an important consideration in medical applications. Discussion Healthcare datasets present several challenges for AutoML tools, including large sample size, high imbalance, and limitations in the available features. Improvements in scalability, combinations of imbalance-learning resampling and ensemble approaches, and curated feature selection are possible next steps to achieve better performance. Conclusion Among the three explored, no AutoML tool consistently outperforms the rest in terms of predictive performance. The performances of the models in this study suggest that there may be room for improvement in handling medical claims data. Finally, selection of the optimal prediction threshold should be guided by the specific practical application.

Abstract The opioid epidemic continues to contribute to loss of life through overdose and significant social and economic burdens. Many individuals who develop problematic opioid use (POU) do so after being exposed to prescribed opioid analgesics. Therefore, it is important to accurately identify and classify risk factors for POU. In this review, we discuss the etiology of POU and highlight novel approaches to identifying its risk factors. These approaches include the application of polygenic risk scores (PRS) and diverse machine learning (ML) algorithms used in tandem with data from electronic health records (EHR), clinical notes, patient demographics, and digital footprints. The implementation and synergy of these types of data and approaches can greatly assist in reducing the incidence of POU and opioid-related mortality by increasing the knowledge base of patient-related risk factors, which can help to improve prescribing practices for opioid analgesics.

Abstract Integrating data across institutions can improve learning efficiency. To integrate data efficiently while protecting privacy, we propose A one-shot, summary-statistics-based, Distributed Algorithm for fitting Penalized (ADAP) regression models across multiple datasets. ADAP utilizes patient-level data from a lead site and incorporates the first-order (ADAP1) and second-order gradients (ADAP2) of the objective function from collaborating sites to construct a surrogate objective function at the lead site, where model fitting is then completed with proper regularizations applied. We evaluate the performance of the proposed method using both simulation and a real-world application to study risk factors for opioid use disorder (OUD) using 15,000 patient data from the OneFlorida Clinical Research Consortium. Our results show that ADAP performs nearly the same as the pooled estimator but achieves higher estimation accuracy and better variable selection than the local and average estimators. Moreover, ADAP2 successfully handles heterogeneity in covariate distributions.

Abstract Given the growing number of prediction algorithms developed to predict COVID-19 mortality, we evaluated the transportability of a mortality prediction algorithm using a multi-national network of healthcare systems. We predicted COVID-19 mortality using baseline commonly measured laboratory values and standard demographic and clinical covariates across healthcare systems, countries, and continents. Specifically, we trained a Cox regression model with nine measured laboratory test values, standard demographics at admission, and comorbidity burden pre-admission. These models were compared at site, country, and continent level. Of the 39,969 hospitalized patients with COVID-19 (68.6% male), 5717 (14.3%) died. In the Cox model, age, albumin, AST, creatine, CRP, and white blood cell count are most predictive of mortality. The baseline covariates are more predictive of mortality during the early days of COVID-19 hospitalization. Models trained at healthcare systems with larger cohort size largely retain good transportability performance when porting to different sites. The combination of routine laboratory test values at admission along with basic demographic features can predict mortality in patients hospitalized with COVID-19. Importantly, this potentially deployable model differs from prior work by demonstrating not only consistent performance but also reliable transportability across healthcare systems in the US and Europe, highlighting the generalizability of this model and the overall approach.

Abstract Background Gene set enrichment analysis (GSEA) uses gene-level univariate associations to identify gene set-phenotype associations for hypothesis generation and interpretation. We propose that GSEA can be adapted to incorporate SNP and gene-level interactions. To this end, gene scores are derived by Relief-based feature importance algorithms that efficiently detect both univariate and interaction effects (MultiSURF) or exclusively interaction effects (MultiSURF*). We compare these interaction-sensitive GSEA approaches to traditional χ 2 rankings in simulated genome-wide array data, and in a target and replication cohort of congenital heart disease patients with conotruncal defects (CTDs). Results In the simulation study and for both CTD datasets, both Relief-based approaches to GSEA captured more relevant and significant gene ontology terms compared to the univariate GSEA. Key terms and themes of interest include cell adhesion, migration, and signaling. A leading edge analysis highlighted semaphorins and their receptors, the Slit-Robo pathway, and other genes with roles in the secondary heart field and outflow tract development. Conclusions Our results indicate that interaction-sensitive approaches to enrichment analysis can improve upon traditional univariate GSEA. This approach replicated univariate findings and identified additional and more robust support for the role of the secondary heart field and cardiac neural crest cell migration in the development of CTDs.

Abstract Background Despite decades of research and established treatment strategies, hypertension remains a prevalent and intractable problem at the population level. Yoga, a lifestyle-based practice, has demonstrated antihypertensive effects in clinical trial settings, but little is known about its effectiveness in the real world. Here, we use electronic health records to investigate the antihypertensive effects of yoga as used by patients in their daily lives. Methods A retrospective, observational case-control study of 1815 records among 1355 yoga exposed patients and 40,326 records among 8682 yoga non-exposed patients collected between 2006 and 2016 from a regional academic health system. Linear mixed-effects models were used to estimate the average treatment effect of yoga on systolic and diastolic blood pressures. Mixed effects logistic regression models were used to calculate odds ratios for yoga use and four blood pressure categories: normal, elevated, stage I, and stage II hypertension. Results Yoga patients are predominantly white (88.0%) and female (87.8%) with median age 46 years (IQR 32–57) who use yoga one time per week (62.3%). Yoga is associated with lower systolic (− 2.8 mmHg, standard error 0.6; p

Summary: Background: While acute kidney injury (AKI) is a common complication in COVID-19, data on post-AKI kidney function recovery and the clinical factors associated with poor kidney function recovery is lacking. Methods: A retrospective multi-centre observational cohort study comprising 12,891 hospitalized patients aged 18 years or older with a diagnosis of SARS-CoV-2 infection confirmed by polymerase chain reaction from 1 January 2020 to 10 September 2020, and with at least one serum creatinine value 1–365 days prior to admission. Mortality and serum creatinine values were obtained up to 10 September 2021. Findings: Advanced age (HR 2.77, 95%CI 2.53–3.04, p

Abstract The genetic basis of phenotypic variation across populations has not been well explained for most traits. Several factors may cause disparities, from variation in environments to divergent population genetic structure. We hypothesized that a population-level polygenic risk score (PRS) can explain phenotypic variation among geographic populations based solely on risk allele frequencies. We applied a population-specific PRS (psPRS) to 26 populations from the 1000 Genomes to four phenotypes: lactase persistence (LP), melanoma, multiple sclerosis (MS) and height. Our models assumed additive genetic architecture among the polymorphisms in the psPRSs, as is convention. Linear psPRSs explained a significant proportion of trait variance ranging from 0.32 for height in men to 0.88 for melanoma. The best models for LP and height were linear, while those for melanoma and MS were nonlinear. As not all variants in a PRS may confer similar, or even any, risk among diverse populations, we also filtered out SNPs to assess whether variance explained was improved using psPRSs with fewer SNPs. Variance explained usually improved with fewer SNPs in the psPRS and was as high as 0.99 for height in men using only 548 of the initial 4208 SNPs. That reducing SNPs improves psPRSs performance may indicate that missing heritability is partially due to complex architecture that does not mandate additivity, undiscovered variants or spurious associations in the databases. We demonstrated that PRS-based analyses can be used across diverse populations and phenotypes for population prediction and that these comparisons can identify the universal risk variants.

Abstract Environmental disasters are anthropogenic catastrophic events that affect health. Famous disasters include the Seveso disaster and the Fukushima-Daiichi nuclear meltdown, which had disastrous health consequences. Traditional methods for studying environmental disasters are costly and time-intensive. We propose the use of electronic health records (EHR) and informatics methods to study the health effects of emergent environmental disasters in a cost-effective manner. An emergent environmental disaster is exposure to perfluoroalkyl substances (PFAS) in the Philadelphia area. Penn Medicine (PennMed) comprises multiple hospitals and facilities within the Philadelphia Metropolitan area, including over three thousand PFAS-exposed women living in one of the highest PFAS exposure areas nationwide. We developed a high-throughput method that utilizes only EHR data to evaluate the disease risk in this heavily exposed population. We replicated all five disease/conditions implicated by PFAS exposure, including hypercholesterolemia, thyroid disease, proteinuria, kidney disease and colitis, either directly or via closely related diagnoses. Using EHRs coupled with informatics enables the health impacts of environmental disasters to be more easily studied in large cohorts versus traditional methods that rely on interviews and expensive serum-based testing. By reducing cost and increasing the diversity of individuals included in studies, we can overcome many of the hurdles faced by previous studies, including a lack of racial and ethnic diversity. This proof-of-concept study confirms that EHRs can be used to study human health and disease impacts of environmental disasters and produces equivalent disease-exposure knowledge to prospective epidemiology studies while remaining cost-effective.

Abstract Conservation machine learning conserves models across runs, users, and experiments—and puts them to good use. We have previously shown the merit of this idea through a small-scale preliminary experiment, involving a single dataset source, 10 datasets, and a single so-called cultivation method—used to produce the final ensemble. In this paper, focusing on classification tasks, we perform extensive experimentation with conservation random forests, involving 5 cultivation methods (including a novel one introduced herein—lexigarden), 6 dataset sources, and 31 datasets. We show that significant improvement can be attained by making use of models we are already in possession of anyway, and envisage the possibility of repositories of models (not merely datasets, solutions, or code), which could be made available to everyone, thus having conservation live up to its name, furthering the cause of data and computational science.

Abstract Background Non-additive interactions among genes are frequently associated with a number of phenotypes, including known complex diseases such as Alzheimer’s, diabetes, and cardiovascular disease. Detecting interactions requires careful selection of analytical methods, and some machine learning algorithms are unable or underpowered to detect or model feature interactions that exhibit non-additivity. The Random Forest method is often employed in these efforts due to its ability to detect and model non-additive interactions. In addition, Random Forest has the built-in ability to estimate feature importance scores, a characteristic that allows the model to be interpreted with the order and effect size of the feature association with the outcome. This characteristic is very important for epidemiological and clinical studies where results of predictive modeling could be used to define the future direction of the research efforts. An alternative way to interpret the model is with a permutation feature importance metric which employs a permutation approach to calculate a feature contribution coefficient in units of the decrease in the model’s performance and with the Shapely additive explanations which employ cooperative game theory approach. Currently, it is unclear which Random Forest feature importance metric provides a superior estimation of the true informative contribution of features in genetic association analysis. Results To address this issue, and to improve interpretability of Random Forest predictions, we compared different methods for feature importance estimation in real and simulated datasets with non-additive interactions. As a result, we detected a discrepancy between the metrics for the real-world datasets and further established that the permutation feature importance metric provides more precise feature importance rank estimation for the simulated datasets with non-additive interactions. Conclusions By analyzing both real and simulated data, we established that the permutation feature importance metric provides more precise feature importance rank estimation in the presence of non-additive interactions.

Thus far, pleiotropy analysis using individual-level Electronic Health Records data has been limited to data from one site. Here, the authors introduce Sum-Share, a method designed to efficiently and losslessly integrate EHR and genetic data from multiple sites to perform pleiotropy analysis.

Abstract Background A typical task in bioinformatics consists of identifying which features are associated with a target outcome of interest and building a predictive model. Automated machine learning (AutoML) systems such as the Tree-based Pipeline Optimization Tool (TPOT) constitute an appealing approach to this end. However, in biomedical data, there are often baseline characteristics of the subjects in a study or batch effects that need to be adjusted for in order to better isolate the effects of the features of interest on the target. Thus, the ability to perform covariate adjustments becomes particularly important for applications of AutoML to biomedical big data analysis. Results We developed an approach to adjust for covariates affecting features and/or target in TPOT. Our approach is based on regressing out the covariates in a manner that avoids ‘leakage’ during the cross-validation training procedure. We describe applications of this approach to toxicogenomics and schizophrenia gene expression data sets. The TPOT extensions discussed in this work are available at https://github.com/EpistasisLab/tpot/tree/v0.11.1-resAdj . Conclusions In this work, we address an important need in the context of AutoML, which is particularly crucial for applications to bioinformatics and medical informatics, namely covariate adjustments. To this end we present a substantial extension of TPOT, a genetic programming based AutoML approach. We show the utility of this extension by applications to large toxicogenomics and differential gene expression data. The method is generally applicable in many other scenarios from the biomedical field.

Abstract Data science has emerged as an important discipline in the era of big data and biological and biomedical data mining. As such, we have seen a rapid increase in the number of data science departments, research centers, and schools. We review here ten important leadership roles for a successful academic data science chair, director, or dean. These roles include the visionary, executive, cheerleader, manager, enforcer, subordinate, educator, entrepreneur, mentor, and communicator. Examples specific to leadership in data science are given for each role.

Abstract We leveraged the largely untapped resource of electronic health record data to address critical clinical and epidemiological questions about Coronavirus Disease 2019 (COVID-19). To do this, we formed an international consortium (4CE) of 96 hospitals across five countries ( www.covidclinical.net ). Contributors utilized the Informatics for Integrating Biology and the Bedside (i2b2) or Observational Medical Outcomes Partnership (OMOP) platforms to map to a common data model. The group focused on temporal changes in key laboratory test values. Harmonized data were analyzed locally and converted to a shared aggregate form for rapid analysis and visualization of regional differences and global commonalities. Data covered 27,584 COVID-19 cases with 187,802 laboratory tests. Case counts and laboratory trajectories were concordant with existing literature. Laboratory tests at the time of diagnosis showed hospital-level differences equivalent to country-level variation across the consortium partners. Despite the limitations of decentralized data generation, we established a framework to capture the trajectory of COVID-19 disease in patients and their response to interventions.

Abstract The coronavirus disease 2019 (COVID-19) pandemic has had a significant impact on population health and wellbeing. Biomedical informatics is central to COVID-19 research efforts and for the delivery of healthcare for COVID-19 patients. Critical to this effort is the participation of informaticians who typically work on other basic science or clinical problems. The goal of this editorial is to highlight some examples of COVID-19 research areas that could benefit from informatics expertise. Each research idea summarizes the COVID-19 application area, followed by an informatics methodology, approach, or technology that could make a contribution. It is our hope that this piece will motivate and make it easy for some informaticians to adopt COVID-19 research projects.

We present AddGBoost, a gradient boosting-style algorithm, wherein the decision tree is replaced by a succession of (possibly) stronger learners, which are optimized via a state-of-the-art hyperparameter optimizer. Through experiments over 90 regression datasets we show that AddGBoost emerges as the top performer for 33% (with 2 stages) up to 42% (with 5 stages) of the datasets, when compared with seven well-known machine-learning algorithms: KernelRidge, LassoLars, SGDRegressor, LinearSVR, DecisionTreeRegressor, HistGradientBoostingRegressor, and LGBMRegressor.

Background: Accumulating evidence has suggested that there is a positive association between asthma and cardiovascular diseases (CVDs), implying a common architecture between them. However, the shared genetic architecture and causality of asthma and CVDs remain unclear.Methods: Based on the genome-wide association study (GWAS) summary statistics of recently published studies, our study examined the genetic correlation, shared genetic variants, and causal relationship between asthma (N = 127,669) and CVDs (N = 86,995–521,612). Statistical methods included high-definition likelihood (HDL), cross-trait meta-analyses of large-scale GWAS, transcriptome-wide association studies (TWAS), and Mendelian randomization (MR).Results: First, we observed a significant genetic correlation between asthma and heart failure (HF) (Rg = 0.278, P = 5 × 10−4). Through cross-trait analyses, we identified a total of 145 shared loci between asthma and HF. Fifteen novel loci were not previously reported for association with either asthma or HF. Second, we mapped these 145 loci to a total of 99 genes whose expressions are enriched in a broad spectrum of tissues, including the seminal vesicle, tonsil, appendix, spleen, skin, lymph nodes, breast, cervix and uterus, skeletal muscle, small intestine, lung, prostate, cardiac muscle, and liver. TWAS analysis identified five significant genes shared between asthma and HF in tissues from the hemic and immune system, digestive system, integumentary system, and nervous system. GSDMA, GSDMB, and ORMDL3 are statistically independent genetic effects from all shared TWAS genes between asthma and HF. Third, through MR analysis, genetic liability to asthma was significantly associated with heart failure at the Bonferroni-corrected significance level. The odds ratio (OR) is 1.07 [95% confidence interval (CI): 1.03–1.12; p = 1.31 × 10−3] per one-unit increase in loge odds of asthma.Conclusion: These findings provide strong evidence of genetic correlations and causal relationship between asthma and HF, suggesting a shared genetic architecture for these two diseases.

Abstract Background Hepatitis C affects about 3 % of the world’s population. In the United States, about 3.5 million have chronic hepatitis C, and it is the leading cause of liver cancer and the most common indication for liver transplantation. In the last decades, new advances in therapy have substantially increased the cure rate of hepatitis C to more than 95% with the use of antiviral agents. However, drug safety of the new treatments remains one of the major concerns. Data from the US Food and Drug Administration (FDA) Adverse Event Reporting System (FAERS) and the Electronic Health Record (EHR) systems provide crucial post-market information to evaluate drug safety. Currently, quantitative evidence of drug safety of hepatitis C treatments based on post-market data are still limited, and there is also a lack of a standard statistical procedure to systematically compare drug safety across multiple drugs using FAERS and EHR. Method In this study, we presented a statistical procedure to compare the difference in adverse events (AE) across multiple hepatitis C drugs using data from FAERS and EHR, and to assess the consistency of results from two data bases. Through three major steps, including descriptive comparison, testing for difference among groups, and quantification of association, the proposed method can provide a quantitative comparison on safety of multiple drugs. Specifically, we compared drugs that were approved by FDA to treat hepatitis C before 2011versus those approved after 2013. We used spontaneous AE reports submitted between 2004 to 2015 from FAERS data base and medical records between 1999 to 2015 from the Cerner health facts data base to estimate and compare the rate of AE after drug use. Result We studied 30 most frequently reported AEs after treatment of hepatitis C, comparing the difference between drugs approved before 2011versus those approved after 2013. Our results showed that there was difference in rate of AE between the two groups of treatment. We reported the AEs that have significant statistical difference, and estimate the difference attributable to variation of age and gender between the two groups of drug users. Our findings are consistent with results in existing literature. Moreover, we compared the results obtained from FAERS data and EHR data, and evaluated the consistency of evidence. Conclusion The proposed procedure is a general and standardized pipeline that can be used to compare and visualize drug safety among multiple drugs to support regulatory decision-makings using post-market data. We showed that there was statistically significant difference in AE rates between the new and old therapies for hepatitis C. We showed that both FAERS and EHR contained large information for research of post-market drug safety, but each has its own strength and limitations. Cautions should be taken when combining evidence from the two data resources and there is a need of more sophisticated informatics and statistical tools for evidence synthesis.

Abstract Background The principal line of investigation in Genome Wide Association Studies (GWAS) is the identification of main effects, that is individual Single Nucleotide Polymorphisms (SNPs) which are associated with the trait of interest, independent of other factors. A variety of methods have been proposed to this end, mostly statistical in nature and differing in assumptions and type of model employed. Moreover, for a given model, there may be multiple choices for the SNP genotype encoding. As an alternative to statistical methods, machine learning methods are often applicable. Typically, for a given GWAS, a single approach is selected and utilized to identify potential SNPs of interest. Even when multiple GWAS are combined through meta-analyses within a consortium, each GWAS is typically analyzed with a single approach and the resulting summary statistics are then utilized in meta-analyses. Results In this work we use as case studies a Type 2 Diabetes (T2D) and a breast cancer GWAS to explore a diversity of applicable approaches spanning different methods and encoding choices. We assess similarity of these approaches based on the derived ranked lists of SNPs and, for each GWAS, we identify a subset of representative approaches that we use as an ensemble to derive a union list of top SNPs. Among these are SNPs which are identified by multiple approaches as well as several SNPs identified by only one or a few of the less frequently used approaches. The latter include SNPs from established loci and SNPs which have other supporting lines of evidence in terms of their potential relevance to the traits. Conclusions Not every main effect analysis method is suitable for every GWAS, but for each GWAS there are typically multiple applicable methods and encoding options. We suggest a workflow for a single GWAS, extensible to multiple GWAS from consortia, where representative approaches are selected among a pool of suitable options, to yield a more comprehensive set of SNPs, potentially including SNPs that would typically be missed with the most popular analyses, but that could provide additional valuable insights for follow-up.

Abstract The Genetic Analysis Workshop (GAW) presents an opportunity to collaboratively evaluate methodology relevant to current issues in genetic epidemiology. The GAW20 data combine real clinical trial data with fictitious epigenetic drug response endpoints. Considering the evidence suggesting that networks of interactions between many genes underlie complex phenotypes, we utilize differential methylation status to identify a relevant gene set for enrichment analysis and use this to infer potential biological function underlying drug response. We highlight the pertinence of considering the potential for widespread epistatic interactions in the absence of main effects, and present evidence of epistasis between single-nucleotide polymorphisms (SNPs) on the two RNA demethylases FTO and ALKBH5.

Abstract Background Machine learning methods and conventions are increasingly employed for the analysis of large, complex biomedical data sets, including genome-wide association studies (GWAS). Reproducibility of machine learning analyses of GWAS can be hampered by biological and statistical factors, particularly so for the investigation of non-additive genetic interactions. Application of traditional cross validation to a GWAS data set may result in poor consistency between the training and testing data set splits due to an imbalance of the interaction genotypes relative to the data as a whole. We propose a new cross validation method, proportional instance cross validation (PICV), that preserves the original distribution of an independent variable when splitting the data set into training and testing partitions. Results We apply PICV to simulated GWAS data with epistatic interactions of varying minor allele frequencies and prevalences and compare performance to that of a traditional cross validation procedure in which individuals are randomly allocated to training and testing partitions. Sensitivity and positive predictive value are significantly improved across all tested scenarios for PICV compared to traditional cross validation. We also apply PICV to GWAS data from a study of primary open-angle glaucoma to investigate a previously-reported interaction, which fails to significantly replicate; PICV however improves the consistency of testing and training results. Conclusions Application of traditional machine learning procedures to biomedical data may require modifications to better suit intrinsic characteristics of the data, such as the potential for highly imbalanced genotype distributions in the case of epistasis detection. The reproducibility of genetic interaction findings can be improved by considering this variable imbalance in cross validation implementation, such as with PICV. This approach may be extended to problems in other domains in which imbalanced variable distributions are a concern.

Abstract Background Machine learning methods have gained popularity and practicality in identifying linear and non-linear effects of variants associated with complex disease/traits. Detection of epistatic interactions still remains a challenge due to the large number of features and relatively small sample size as input, thus leading to the so-called “short fat data” problem. The efficiency of machine learning methods can be increased by limiting the number of input features. Thus, it is very important to perform variable selection before searching for epistasis. Many methods have been evaluated and proposed to perform feature selection, but no single method works best in all scenarios. We demonstrate this by conducting two separate simulation analyses to evaluate the proposed collective feature selection approach. Results Through our simulation study we propose a collective feature selection approach to select features that are in the “union” of the best performing methods. We explored various parametric, non-parametric, and data mining approaches to perform feature selection. We choose our top performing methods to select the union of the resulting variables based on a user-defined percentage of variants selected from each method to take to downstream analysis. Our simulation analysis shows that non-parametric data mining approaches, such as MDR, may work best under one simulation criteria for the high effect size (penetrance) datasets, while non-parametric methods designed for feature selection, such as Ranger and Gradient boosting, work best under other simulation criteria. Thus, using a collective approach proves to be more beneficial for selecting variables with epistatic effects also in low effect size datasets and different genetic architectures. Following this, we applied our proposed collective feature selection approach to select the top 1% of variables to identify potential interacting variables associated with Body Mass Index (BMI) in ~ 44,000 samples obtained from Geisinger’s MyCode Community Health Initiative (on behalf of DiscovEHR collaboration). Conclusions In this study, we were able to show that selecting variables using a collective feature selection approach could help in selecting true positive epistatic variables more frequently than applying any single method for feature selection via simulation studies. We were able to demonstrate the effectiveness of collective feature selection along with a comparison of many methods in our simulation analysis. We also applied our method to identify non-linear networks associated with obesity.

Abstract Evolutionary computation (EC) has been widely applied to biological and biomedical data. The practice of EC involves the tuning of many parameters, such as population size, generation count, selection size, and crossover and mutation rates. Through an extensive series of experiments over multiple evolutionary algorithm implementations and 25 problems we show that parameter space tends to be rife with viable parameters, at least for the problems studied herein. We discuss the implications of this finding in practice for the researcher employing EC.

Abstract Background The selection, development, or comparison of machine learning methods in data mining can be a difficult task based on the target problem and goals of a particular study. Numerous publicly available real-world and simulated benchmark datasets have emerged from different sources, but their organization and adoption as standards have been inconsistent. As such, selecting and curating specific benchmarks remains an unnecessary burden on machine learning practitioners and data scientists. Results The present study introduces an accessible, curated, and developing public benchmark resource to facilitate identification of the strengths and weaknesses of different machine learning methodologies. We compare meta-features among the current set of benchmark datasets in this resource to characterize the diversity of available data. Finally, we apply a number of established machine learning methods to the entire benchmark suite and analyze how datasets and algorithms cluster in terms of performance. From this study, we find that existing benchmarks lack the diversity to properly benchmark machine learning algorithms, and there are several gaps in benchmarking problems that still need to be considered. Conclusions This work represents another important step towards understanding the limitations of popular benchmarking suites and developing a resource that connects existing benchmarking standards to more diverse and efficient standards in the future.

Centralized infrastructure to support analyses involving complexity beyond genome-wide association studies is broadly needed. Here, Ritchie and colleagues develop PLATO, a software tool to process and integrate various methods for this task.

Abstract Background The genetic etiology of human lipid quantitative traits is not fully elucidated, and interactions between variants may play a role. We performed a gene-centric interaction study for four different lipid traits: low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), total cholesterol (TC), and triglycerides (TG). Results Our analysis consisted of a discovery phase using a merged dataset of five different cohorts (n = 12,853 to n = 16,849 depending on lipid phenotype) and a replication phase with ten independent cohorts totaling up to 36,938 additional samples. Filters are often applied before interaction testing to correct for the burden of testing all pairwise interactions. We used two different filters: 1. A filter that tested only single nucleotide polymorphisms (SNPs) with a main effect of p