Knowledge-guided learning methods for integrative analysis of multi-omics data.

Integrative analysis of multi-omics data has the potential to yield valuable and comprehensive insights into the molecular mechanisms underlying complex diseases such as cancer and Alzheimer's disease. However, a number of analytical challenges complicate multi-omics data integration. For instance, -omics data are usually high-dimensional, and sample sizes in multi-omics studies tend to be modest. Furthermore, when genes in an important pathway have relatively weak signal, it can be difficult to detect them individually. There is a growing body of literature on knowledge-guided learning methods that can address these challenges by incorporating biological knowledge such as functional genomics and functional proteomics into multi-omics data analysis. These methods have been shown to outperform their counterparts that do not utilize biological knowledge in tasks including prediction, feature selection, clustering, and dimension reduction. In this review, we survey recently developed methods and applications of knowledge-guided multi-omics data integration methods and discuss future research directions.

Extremely sparse models of linkage disequilibrium in ancestrally diverse association studies.

Linkage disequilibrium (LD) is the correlation among nearby genetic variants. In genetic association studies, LD is often modeled using large correlation matrices, but this approach is inefficient, especially in ancestrally diverse studies. In the present study, we introduce LD graphical models (LDGMs), which are an extremely sparse and efficient representation of LD. LDGMs are derived from genome-wide genealogies; statistical relationships among alleles in the LDGM correspond to genealogical relationships among haplotypes. We published LDGMs and ancestry-specific LDGM precision matrices for 18 million common variants (minor allele frequency >1%) in five ancestry groups, validated their accuracy and demonstrated order-of-magnitude improvements in runtime for commonly used LD matrix computations. We implemented an extremely fast multiancestry polygenic prediction method, BLUPx-ldgm, which performs better than a similar method based on the reference LD correlation matrix. LDGMs will enable sophisticated methods that scale to ancestrally diverse genetic association data across millions of variants and individuals.

Shared components of heritability across genetically correlated traits.

Most disease-associated genetic variants are pleiotropic, affecting multiple genetically correlated traits. Their pleiotropic associations can be mechanistically informative: if many variants have similar patterns of association, they may act via similar pleiotropic mechanisms, forming a shared component of heritability. We developed pleiotropic decomposition regression (PDR) to identify shared components and their underlying genetic variants. We validated PDR on simulated data and identified limitations of existing methods in recovering the true components. We applied PDR to three clusters of five to six traits genetically correlated with coronary artery disease (CAD), asthma, and type II diabetes (T2D), producing biologically interpretable components. For CAD, PDR identified components related to BMI, hypertension, and cholesterol, and it clarified the relationship among these highly correlated risk factors. We assigned variants to components, calculated their posterior-mean effect sizes, and performed out-of-sample validation. Our posterior-mean effect sizes pool statistical power across traits and substantially boost the correlation (r2) between true and estimated effect sizes (compared with the original summary statistics) by 94% and 70% for asthma and T2D out of sample, respectively, and by a predicted 300% for CAD.

Digital signatures for early traumatic brain injury outcome prediction in the intensive care unit.

Traumatic brain injury (TBI) is a leading neurological cause of death and disability across the world. Early characterization of TBI severity could provide a window for therapeutic intervention and contribute to improved outcome. We hypothesized that granular electronic health record data available in the first 24 h following admission to the intensive care unit (ICU) can be used to differentiate outcomes at discharge. Working from two ICU datasets we focused on patients with a primary admission diagnosis of TBI whose length of stay in ICU was ≥ 24 h (N = 1689 and 127). Features derived from clinical, laboratory, medication, and physiological time series data in the first 24 h after ICU admission were used to train elastic-net regularized Generalized Linear Models for the prediction of mortality and neurological function at ICU discharge. Model discrimination, determined by area under the receiver operating characteristic curve (AUC) analysis, was 0.903 and 0.874 for mortality and neurological function, respectively. Model performance was successfully validated in an external dataset (AUC 0.958 and 0.878 for mortality and neurological function, respectively). These results demonstrate that computational analysis of data routinely collected in the first 24 h after admission accurately and reliably predict discharge outcomes in ICU stratum TBI patients.