Unlocking the future of complex human diseases prediction: multi-omics risk score breakthrough

Abstract

Precise prediction of the risk of acquiring complex human diseases using genomic data has gained a considerable traction among clinicians, medical geneticists and researchers, particularly in this era of next generation sequencing. Multi-omics methods utilize various high-throughput screening technologies such as genomics (GWAS), DNA methylomics, metagenomics, transcriptomics, proteomics, metabolomics, and many others which play a crucial role in advancing the understanding of human diseases (Figure 1). These diverse multiomics indicators create a comprehensive framework, yielding significant insights into future health status predictions. The polygenic risk scores (PRS) -a calculation of a person's genetic predisposition to a trait or disease based on their genotype from pertinent genome-wide association study (GWAS) findings (Choi et al., 2020)-as well as methylation risk scores (MRS) -a linear combination of CpG (5'-C-phosphate-G-3') methylation (covalent attachment of a methyl group onto the cytosine residue of DNA) states (Thompson et al., 2022)-have shown promise in predicting complex human diseases accurately (Liu et al., 2024). However, their translation into clinical care is yet to be realized. Several efforts have been made to improve their accuracy in predicting complex human diseases, such as increasing diversity in the genetic training databases, such as the All of Us Research Program, and including conventional risk factors in the PRS model (Liu et al., 2024). In the realm of predictive medicine, conventional risk factors span socio-demographic elements like age and sex, alongside anthropometric data such as body mass index (BMI) and crucial clinical measures, including blood pressure, lipid profiles, kidney and liver function tests, and other key biomarkers such as glycated hemoglobin (HbA1c). These conventional risk factors intertwine with lifestyle choices, behaviors, and environment.Figure 1 shows how Multi-omics risk score is constructed from various high-throughput screening technologies such as genomics (GWAS), DNA methylomics, metagenomics, transcriptomics, proteomics and metabolomics to precisely predict and advancing the understanding of human diseases.The advent of GWAS, methylome-wide association studies (MWAS), and transcriptome-wide association studies (TWAS) have propelled genetic research forward by leaps and bounds, enabling the genotyping, methylation typing, and transcriptome analysis of millions of human samples. Through this vast endeavor, researchers have extracted genetic variants (Wu et al., 2021) and methylation patterns intricately linked to disease susceptibility across the human genome. The genetic variants and methylation patterns serve as the building blocks for constructing PRS (Choi et al., 2020) and MRS tailored to predict complex diseases in individuals based on their unique architecture. The efficacy and clinical potential of these tools shine brightly, offering invaluable insights into risk prediction for a plethora of common complex human diseases including cardiovascular diseases, cancers, diabetes mellitus, Alzheimer's disease, and ankylosing spondylitis (Cappozzo et al., 2022). They represent transformative applications in the arsenal of personalized medicine, promising to revolutionize healthcare by unlocking more secrets hidden within our genomes.The immense potential of genome-wide genotyping arrays lies in their ability to serve as a cost-effective approach capable of generating hundreds of PRSs. This groundbreaking technology is now undergoing rigorous evaluation in clinical studies across global healthcare systems. The allure of PRSs as predictive tools resonates profoundly, offering a glimpse into a future where personalized healthcare is not just a dream, but a tangible reality poised to transform medical practice. The full clinical potential of PRS and MRS remains largely untapped (Martin et al., 2019). This reality is especially pronounced in populations with high genetic diversity, diminished linkage blocks, and historical under-representation in genome databases, such as those of sub-Saharan African descent. The journey towards widespread clinical implementation of PRS is still in its infancy, with considerable challenges to overcome. Yet, with determination and concerted effort, bridging these gaps holds the key to unlocking the transformative power of PRS and MRS in diverse populations worldwide. Of note, there are concerted efforts such as the All of Us Research Program (Bick et al., 2024), Human Heredity and Health Africa (H3Africa), and others, to increase the representation of historically under-represented populations in the global genome databases to leverage this disparity. Ultimately the quantity and quality of data to compute PRS and MRS are escalating.The advent of multi-omics technologies and accrued data thereof in recent era suggest the feasibility of measuring and combining various omics data and cellular factors. This enables the creation of multi-omics risk scores (MoRS) with enhanced predictability for complex diseases (Liu et al., 2024). PRS integrated with multi-omics data analyses, including metagenomics, epigenomics, and transcriptomics have revealed potential biomarkers and ultimately improved predictability for several prevalent age-related conditions like heart disease, diabetes, dementia, and various cancers (Liu et al., 2024).The human gut microbiota, which refers to the collection of microorganisms residing in someone's gastrointestinal tract, has been implicated in numerous common diseases (Chen et al., 2024;Huang et al., 2024). Specific microbial signatures in the gut have been linked to mortality and the development of diseases such as type 2 Diabetes (T2D), liver issues, and respiratory diseases among the general population (Liu et al., 2024). This suggests that the composition of the gut microbiome could potentially aid in predicting disease risk. It is worth noting that while GWAS has shed light on the genetic underpinnings of the gut microbiome, it is evident that the heritability of the gut microbiome is relatively low. Furthermore, similarities in the gut microbiome across generations are primarily associated with living in the same household rather than genetic factors. Recent studies have highlighted the association of various omics data such as gut metagenomics, DNA methylome data from epigenome-wide association studies and transcriptomics data with complex human diseases (Wu et al., 2021;Liu et al., 2024).Recent research indicates that PRS models alone demonstrate superior predictive power compared to traditional risk factors (Liu et al., 2024). Furthermore, integrating MRS and transcriptomics data into the PRS showed a substantial improvement in prediction of prostate cancer (Wu et al., 2021). However, when conventional risk factors are incorporated into the PRS model, performance improves (Liu et al., 2024). Moreover, integrating both conventional risk factors and additional omics data such as from gut metagenomics into the PRS model significantly enhances its predictive performance for complex human diseases (Liu et al., 2024). Therefore, these studies demonstrated that the inclusion of other omics data from gut metagenomics, DNA methylomics and transcriptomics have shown a promise to improve the prediction of the incidence of age-related complex human diseases such as coronary artery disease, type 2 diabetes, Alzheimer's disease, and prostate cancer (Wu et al., 2021;Liu et al., 2024).Recent research suggests that studying blood DNA methylation at various CpG sites can serve as a valuable surrogate biomarker for exposure to risk factors, aiding in the prediction of complex human diseases such as cardiovascular diseases and in identifying high-risk populations (Cappozzo et al., 2022). Methylation risk scores (MRS) are typically constructed to model the relationship between methylation at CpG sites and specific traits or diseases through epigenome-wide association studies (EWAS). DNA methylation scores have been effectively utilized to assess an individual's biological age (epigenetic clock) and have been strongly associated with several non-communicable diseases (NCDs) risk factors such as smoking, alcohol consumption, low physical activity, obesity, socioeconomic status, and occupational characteristics. This existing collinearity has made DNA methylation scores a powerful tool for predicting aging-related diseases as well as lifestyle-related diseases such as cancer and cardiovascular diseases (Cappozzo et al., 2022).These epigenetic clocks demonstrate strong predictive capabilities for aging-related diseases and overall mortality. Research indicates that the risk of developing complex human diseases depends on the interaction between host genetic factors, environmental influences, and human behaviors or lifestyles. Incorporating conventional risk factors such as age, sex, smoking, and alcohol consumption into models accounts for human behavior (Levine et al., 2018). Studies, including the one conducted by Liu et al., have demonstrated that including these conventional risk factors improves the predictive ability for complex human diseases (Liu et al., 2024). Moreover, environmental factors, gene-environment interaction, and host lifestyle can be surrogated by epigenetic methylation analysis. Therefore, it is of the essence for the DNA methylomics data to be integrated into PRS models to enhance predictability for complex human diseases. There is a hypothesis suggesting that epigenomics data from DNA methylation might offer better predictive ability than many of the current PRS utilized today (Thompson et al., 2022). Consequently, integrating these multi-omics data into the PRS model could potentially yield the most effective predictive model for complex human diseases. To the best of our knowledge, there are very limited studies reported to integrate the DNA methylomics data into the PRS. Therefore, further investigations are warranted to explore the impact of integrating DNA methylomics data into PRS models for enhancing and predicting the development of complex human diseases.Epigenetic modifications are widely recognized as influential factors in the biological pathways of both communicable diseases and non-communicable complex human diseases like hypertension and cancer with DNA methylation being the most extensively studied. Epigenetics involves the alteration of gene expression without changing the genetic code through processes such as DNA methylation and histone modification. This procedure entails attaching covalently a methyl group to the cytosine base found within sections containing repeated cytosine-guanine bonds, also referred to as CpG islands. When a gene undergoes heavy methylation, it typically remains transcriptionally silent (Irizarry et al., 2009). Environmental factors can trigger significant changes in methylation levels. The methylation patterns found in promoter CpG islands, which are clusters of CpG sites located in gene promoters, hold significant promise as potential biomarkers. They could play crucial roles in disease detection, disease classification, prognosis, and forecasting treatment responses (Ehrlich, 2019).Recent research has revealed compelling links between DNA methylation and fluctuations in blood pressure, cardiovascular ailments, and various other non-communicable diseases. Han et al highlighted the pivotal role of gene-specific DNA methylation in elevating blood pressure, notably concerning factors like angiotensin-converting enzyme, lipid and amino acid metabolism, and impaired glucose metabolism (Han et al., 2016). Richard et al. pinpointed 13 replicated CpG sites, explaining 1.4% and 2.0% of individual differences in systolic and diastolic blood pressure, respectively (Richard et al., 2017). Intriguingly, new findings propose a strong link between DNA methylation and lifestyle choices (like smoking, alcohol intake, and diet), aging, obesity, and gender-all vital risk factors for hypertension. Kim et al. identified an association between DNA methylation in peripheral blood leukocytes and hypertension prevalence hints at the potential of DNA methylation as a high blood pressure biomarker (Kim et al., 2010). This highlights the promise of integrating DNA methylomics data into models to significantly enhance PRS performance.Constructing multi-omics risk scores for complex human diseases typically involves integrating multiple layers of biological data (genomics, methylomics, metagenomics, transcriptomics, proteomics and metabolomics) into a single predictive score that quantifies disease risk. Bioinformatics and computational tools for construction multi-omics risk scores use various statistical, machine learning, and feature selection techniques to identify predictive markers across omics layers and combine them into a single aggregated risk score. Table 1 summarizes the most commonly used analytical tools for multi-omics data integration. By combining these bioinformatics and computational tools and frameworks, researchers can construct multi-omics risk scores that provide comprehensive, predictive insights into complex human disease susceptibility and progression. Ultimately enhancing our understanding of the molecular basis of these complex diseases by leveraging complementary information across multi-omics data. (Euesden et al., 2015).PRS-CS is a Bayesian polygenic risk scoring tool that improves the accuracy of PRS by accounting for linkage disequilibrium via Bayesian regression and continuous shrinkage (CS) priors. This tool allows integrating genomics data with other omics layers, such as transcriptomics data, to build multi-omics risk scores (Ge et al., 2019).A deep learning-based tool that integrates multi-omics data using neural networks to identify predictive biomarkers and generate risk scores. Its architecture can handle complex, non-linear relationships across omics layers, making it well-suited for multi-omics risk prediction (Kang et al., 2022).These machine learning tools use various algorithms to integrate omics data and predict disease risk scores. They support feature selection, classification, and regression to create multi-omics risk prediction models (Zoppi et al., 2021;Ballard et al., 2024).Analysis+)It performs factor analysis to reduce dimensionality across omics datasets, identifying factors that contribute to disease risk. These factors can then be combined to build risk scores for predicting disease phenotypes and clinical outcomes (Argelaguet et al., 2020).CustOmics is a versatile deep-learning-based framework for multi-omics integration, designed for survival and classification tasks. It leverages customizable architectures to integrate data across omics types, particularly in cancer research (Benkirane et al., 2023).PLIER is a tool for dimensionality reduction and feature selection that leverages known pathways to combine multiple omics layers. It can generate interpretable factors used in risk score modeling, making it ideal for multi-omics risk prediction (Mao et al., 2019).CIMLR integrates multi-omics data by learning a consensus clustering across omics layers, useful for stratifying patients and creating disease risk scores. This method is effective in scenarios where disease subtypes must be identified in multi-omics datasets (Ramazzotti et al., 2018).BOI is a Bayesian framework that uses priors based on biological knowledge (e.g., pathway information) to integrate data across omics layers and predict risk. It is particularly effective in combining genomics and epigenomics data to improve disease risk prediction (Fang et al., 2018;Almutiri et al., 2024;Novoloaca et al., 2024).NetDx combines multi-omics data to build predictive patient similarity networks, which can be used to classify patient risk. NetDx allows for the creation of risk scores based on network-derived patient profiles and has been applied in cancer and psychiatric disease risk prediction (Pai et al., 2019).Mergeomics uses network-based integration to link multi-omics markers with pathways and disease-related gene networks. By identifying critical network modules associated with disease risk, Mergeomics aids in building risk scores that combine multi-omics biomarkers (Shu et al., 2016).This end-to-end analysis pipeline provides workflows for analyzing multi-omics data, including RNA-seq, DNAseq, and epigenomics data. OmicsPipe integrates these layers to build disease prediction models and can be adapted to produce multi-omics risk scores (Fisch et al., 2015).A translational research platform that integrates and analyzes multi-omics data, clinical data, and biomarker information. TranSMART includes tools for multi-omics data integration, risk score modeling, and data visualization. While tranSMART itself does not directly compute risk scores, it can help identify biomarker candidates and generate hypotheses about disease risk factors by linking omics data to clinical outcomes (Athey et al., 2013;tranSMART-Foundation/transmart, 2023).Although primarily focused on metabolomics, MetaboAnalyst has multi-omics capabilities, including pathway analysis and functional annotation. It can identify metabolomics and gene expression biomarkers linked to disease risk. Researchers typically use MetaboAnalyst's results in conjunction with statistical or machine learning models to develop personalized risk scores based on identified pathways and biomarkers (Pang et al., 2024).A network and pathway analysis tool that integrates multi-omics data, including genomics, transcriptomics, and proteomics information, to predict disease-related genes. GeneMANIA's network-based approach can support risk score creation based on pathway associations (Mostafavi et al., 2008).An R package that provides multi-omics integration and visualization methods, including PLS-DA, DIABLO, and multivariate factor analysis. It supports building predictive models and risk scores by selecting key features from multiple omics layers (Rohart et al., 2017).ComplexHeatmap A visualization package in R that supports hierarchical clustering and multi-layered heatmaps for multi-omics data. It is commonly used to visualize relationships across omics layers, aiding in feature selection for risk scoring (Gu et al., 2016;Gu, 2022).The main challenges and limitations for multi-omics utility for improving prediction of complex human diseases are underrepresentation of diverse population in the genome databases, due to the fact that collecting multi-omics data is challenging due to the time consuming and costly process, particularly from large scale human genetic studies. As a result, multi-omics data may only be available for a subset of participants in a study limiting its statistical power, generalizability and hence transferability (Hasin et al., 2017;Chen and Furthermore, of among and biological particularly from low the to its clinical utility of multi-omics for Therefore, the concerted efforts are for of multi-omics databases, as well as training among biological in health and in several key limitations the and of multi-omics are a significant in omics data particularly in studies where are in or These between can impact data in omics studies to differences in data by in conditions across of samples. These differences are to the biological or being not can to can from to study such as or or treatment in and including differences in or can also contribute to such as in or across further the which when are at and environmental factors, such as changes in or other conditions or are additional of et al., are particularly to due to in data by or mass DNA may in or and may in of methods, or proteomics and metabolomics can be by differences in or et al., 2024). data analysis its with from differences in analysis or databases, and the treatment of or these of is critical to the and of multi-omics studies et al., 2024). Therefore, is crucial to the of results in multi-omics studies and the risk of are common in multi-omics data and can to outcomes not or These can biological create reduce the and of and the accuracy of analyses, such as classification, or biomarker To this several algorithms have been to multi-omics data integration. However, their and limitations must be based on the type of omics data, performance and specific scenarios selecting an method for use et al., et al., et al., and is crucial to the and of omics data while the impact of on biological As multi-omics data to the of and effective algorithms is to to research and clinical The by et al. provides insights into the visualization and potential for in omics including an of the available algorithms et al., 2024). between individual studies is significant as conditions can across potentially to results et al., et al., omics types, such as transcriptomics, are particularly to and methods, which can Moreover, differences in platform technologies for the same omics type can further contribute to DNA methylation analysis can be using a of for human DNA These the which the measuring the or measuring and the or measuring et al., as well as the offering of and et al., 2021). metabolomics can significantly on it is or challenges in there is the of of risk, which the translation of findings into applications et al., additional challenges data as integrating diverse data with and is challenging and statistical and computational and are also a as multi-omics data integration is high performance and due to the of multi-omics models that them challenging to and limiting their utility in clinical The and future research of multi-omics risk prediction at the fact that integrating multi-omics data within a framework researchers to model pathways and networks that diseases, to into and potential Multi-omics risk prediction has the potential to based on an individual's molecular personalized medicine, and improving This can be used in population screening to identify high-risk individual for particularly for diseases which is critical such as cardiovascular diseases, diabetes mellitus, Furthermore, the multiomics has the potential to be used in understanding disease by identifying the key molecular of disease, it can new and Multi-omics risk prediction holds the promise for enhancing in complex human diseases risk with the potential to to health care towards more As and computational tools to multi-omics integration play an pivotal role in and complex human human complex diseases from a complex between the host host behaviors or and environmental factors. multi-omics data as metagenomics, epigenomics, particularly DNA methylomics, and transcriptomics conventional risk factors into risk score models holds the potential for the predictive performance for age-related complex human diseases compared to models based on PRS. research the integration of multi-omics data is warranted to enhance PRS performance in predicting complex human diseases.

Topics & Concepts