Uncovering Cross-Cohort Molecular Features with Multi-Omics Integration Analysis

Jiang, M.-Z.; Aguet, F.; Ardlie, K.; Chen, J.; Cornell, E.; Cruz, D.; Durda, P.; Gabriel, S. B.; Gerszten, R. E.; Guo, X.; Johnson, C. W.; Kasela, S.; Lange, L. A.; Lappalainen, T.; Liu, Y.; Reiner, A. P.; Smith, J.; Sofer, T.; Taylor, K. D.; Tracy, R. P.; VanDenBerg, D. J.; Wilson, J. G.; Rich, S. S.; Rotter, J. I.; Love, M. I.; Raffield, L. M.; Li, Y.

doi:10.1101/2022.11.10.515908

Home/Archive/bioRxiv

bioRxiv · 10.1101/2022.11.10.515908

Uncovering Cross-Cohort Molecular Features with Multi-Omics Integration Analysis

View original source ↗Export metadata ↓

Abstract

Integrative approaches that simultaneously model multi-omics data have gained increasing popularity because they provide holistic system biology views of multiple or all components in a biological system of interest. Canonical correlation analysis (CCA) is a correlation-based integrative method. It was initially designed to extract latent features shared between two assays by finding the linear combinations of features - referred to as canonical vectors (CVs) - within each assay that achieve maximal across-assay correlation. Sparse multiple CCA (SMCCA), a widely-used derivative of CCA, allows more than two assays but can result in non-orthogonal CVs when applied to high-dimensional data. Here, we incorporated a variation of the Gram-Schmidt (GS) algorithm with SMCCA to improve orthogonality among CVs. Applying our SMCCA-GS method to proteomics and methylomics data from the Multi-Ethnic Study of Atherosclerosis (MESA) and Jackson Heart Study (JHS), we identified strong associations between blood cell counts and protein abundance. This finding suggests that adjustment of blood cell composition should be considered in protein-based association studies. Importantly, CVs obtained from two independent cohorts demonstrate transferability across the cohorts. For example, proteomic CVs learned from JHS explain similar amounts of blood cell count phenotypic variance in MESA, explaining 39.0% ~ 50.0% variation in JHS and 38.9% ~ 49.1% in MESA, similar transferability was observed for other omics-CV-trait pairs. This suggests that biologically meaningful and cohort-agnostic variation is captured by CVs. We further developed Sparse Supervised Multiple CCA (SSMCCA) to allow supervised integration analysis for more than two assays. We anticipate that applying our SMCCA-GS and SSMCCA on various cohorts would help identify cohort-agnostic biologically meaningful relationships between multi-omics data and phenotypic traits. Author SummaryComprehensive understanding of human complex traits may benefit from incorporation of molecular features from multiple biological layers such as genome, epigenome, transcriptome, proteome, and metabolome. CCA is a correlation-based method for multi-omics data which reduces the dimension of each omic assay to several orthogonal components - commonly referred to as canonical vectors (CVs). The widely-used SMCCA method allows effective dimension reduction and integration of multi-omics data, but suffers from potentially highly correlated CVs when applied to high-dimensional omics data. Here, we improve the statistical independence among the CVs by adopting a variation of the GS algorithm. We applied our SMCCA-GS method to proteomic and methylomic data from two cohort studies, MESA and JHS. Our results reveal a pronounced effect of blood cell counts on protein abundance, strongly suggesting blood cell composition adjustment in protein-based association studies may be necessary. Finally, we present SSMCCA which allows supervised CCA analysis for the association between one phenotype of interest and more than two assays. We anticipate that SMCCA-GS would help reveal meaningful system-level factors from biological processes involving features from multiple assays; and SSMCCA would further empower interrogation of these factors for phenotypic traits related to health and diseases.

Explore related subjects

genomics Preprint

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Jiang, M.-Z., Aguet, F., Ardlie, K., Chen, J., Cornell, E., Cruz, D., Durda, P., Gabriel, S. B., Gerszten, R. E., Guo, X., Johnson, C. W., Kasela, S., Lange, L. A., Lappalainen, T., Liu, Y., Reiner, A. P., Smith, J., Sofer, T., Taylor, K. D., Tracy, R. P., VanDenBerg, D. J., Wilson, J. G., Rich, S. S., Rotter, J. I., Love, M. I., Raffield, L. M., Li, Y.. 2022-11-10. Uncovering Cross-Cohort Molecular Features with Multi-Omics Integration Analysis. https://doi.org/10.1101/2022.11.10.515908

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

MaaD-TAD: an all-atom model of the duplicated PRAMEF/HNRNPCL chromatin domain built from sequence homology

In the nucleus, the placement of genes in the 3D fold of DNA affects how they function. Models of this fold use two kinds of data: contact maps constructed from Hi-C, and nucleosome maps showing where DNA is wound around nucleosomes. Over evolution, sections of the genome were copied, creating near-identical regions. Reads from these regions cannot be assigned to one copy and are discarded, leaving "void regions" in both maps. We tested MaaD-TAD, a pipeline that models topologically associating domains (TADs) at the atomic level, on the PRAMEF/HNRNPCL locus on human chromosome 1. This challenging locus spans 652 kb, with only 15% of its nucleosome positions known and most contact data missing. To test our hypothesis that copies with similar sequence lie close in space, i.e. homology as a surrogate for proximity, we constructed a map of sequence homology in place of Hi-C and used CSynth to fold it into a 3D path. We then lengthened the path to fit a chromatin fibre, and refined the model by adding cohesin complexes, which stabilize chromatin structure. Cohesin is a ring-shaped protein complex through which a loop of chromatin is threaded; the loop is commonly anchored by CTCF proteins bound on either leg. Using known databases, we placed CTCF sites in our model and joined three to their copies with cohesin rings; in each pair the sites face each other, as loop anchors do. The final model holds 3,359 nucleosomes and about 55 million atoms. Micro-C, an advanced, higher-resolution form of Hi-C, provided data on two other duplicated regions that support our hypothesis: matching copies touched 1.2 to 1.9 times more often than expected. Our approach provides a novel method to reconstruct DNA regions with absent contact data and, to our knowledge, the first all-atom model of a complete TAD.

genomics↗

bioRxiv2026

Pseudo-observations of longevity improve detection of resilience-associated genes in the Radiation Late Effects Cohort of rhesus macaques.

Non-human primates are an ideal model for interrogating the genetic basis of human lifespan and resilience due to their long lifespan compared to other model organisms, development of similar age-related chronic conditions, and genetic similarity. However, like humans, statistical methods for modeling survival suffer from the challenges of dealing with non-proportionality of hazards, large numbers of living individuals that must be censored, and loss-to-follow-up over studies that may last a decade or more. Here we overcome these challenges through the use of pseudo-observations of Restricted Mean Survival Time as a measure of resilience to the acute challenge of radiation exposure and chronic challenges of aging and a Western-style diet in rhesus macaques (Macaca mulatta) from the Wake Forest Non-Human Primate Radiation Late Effects Cohort. Our whole-exome analysis used Genome-wide Efficient Mixed Model Association (GEMMA) to identify variants associated with longevity, and a subsequent Gene Set Enrichment Analysis (GSEA) highlighted the cadherin and integrin signaling pathways as central to this resilience. As dysregulation of cell adhesion and extracellular matrix signaling is central to the development of fibrosis, we tested the functional relevance of these signals using a targeted Combined Annotation Dependent Depletion (CADD) score-weighted Sequence Kernel Association Test-Optimal (SKAT-O) analysis of GSEA leading-edge genes from both the cadherin and integrin pathways. This linked variant burden in PCDHGA5, PCDHGA3, PCDHGB2, PCDHA11, and PSENEN to the severity of myocardial fibrosis, and in IBSP, PDGFD, and LAMC2 to the severity of pulmonary fibrosis observed at necropsy. Critically, we note that the genetic variants are not caused by the radiation but rather create an underlying risk for disease that is then exacerbated by the acute and persistent stressors challenging these animals' intrinsic resilience. The success of this method in identifying biologically relevant genetic variants underlying survival in animals undergoing a strong challenge to their resilience highlights not only the role of genes predisposing animals to fibrosis in lifespan, but the overall utility of pseudo-observations in overcoming the limitations of survival analyses in humans and other long-lived species.

genomics↗

bioRxiv2026

Segment-dependent phylogenetic resolution reveals geographic and host sampling gaps in Oropouche virus genomes

Background. Oropouche virus (OROV) has been reported in areas beyond its historical Amazonian range since 2023, accompanied by rapid growth in publicly available genomic data. However, the extent to which these data provide reliable phylogenetic and epidemiological resolution, and whether that resolution differs among genome segments, has not been systematically examined. Methods. Nucleotide sequences annotated under Orthobunyavirus oropoucheense were retrieved from NCBI Virus (accessed 13 May 2025) and filtered to retain records at least 70% of the reference length for each segment, yielding 898 S, 831 M, and 832 L sequences. Segments were aligned independently with MAFFT, and maximum-likelihood phylogenies were inferred with IQ-TREE using ModelFinder and ultrafast bootstrap approximation. Trees were annotated with collection date, country, and host in iTOL. Results. Phylogenetic resolution differed among segments. Contemporary Brazilian sequences formed a largely unresolved region in the S segment tree, whereas the M and L trees resolved recent clusters to differing degrees. Resolution did not track segment length: the M tree resolved contemporary structure more strongly than the longer L tree, and total tree length was approximately 3.5 times greater in the M tree despite near-identical taxon sampling. Sequences collected before 2010 were recovered in supported clusters in all three segment trees, although their configuration differed by segment. These sequences were also predominantly non-Brazilian and non-human, preventing collection date, geographic origin, and host from being separated as explanations for the observed clustering. The species-level query also retained three reassortant viruses, which may influence segment-specific topologies; sequences annotated to the United States were Iquitos virus rather than OROV. Two Italian sequences clustered with Cuban sequences across all three trees, concordant with published travel histories. Conclusions. Publicly available sequences contain phylogenetic structure that differs among the S, M, and L segment trees, but interpretation is constrained by uneven temporal, geographic, and host sampling and by the inclusion of reassortant species in a species-level dataset. The concentration of sequences from Brazil and human hosts, together with limited representation of non-human hosts, restricts inference about OROV diversity and ecology. Broader and more balanced genomic surveillance would improve the ability to resolve OROV evolutionary and transmission patterns.

genomics↗