Unsupervised factor integration
Shared latent factors across layers with variance explained per view—an unbiased map of coordinated variation, no labels required.
Core Bioinformatics
Multi-Omics Integration
From separate omics layers to one coherent model. We integrate genomics, transcriptomics, epigenomics, proteomics, and metabolomics—using factor analysis, supervised integration, and network fusion to find the signal that runs across layers—and hand back factors, drivers, and figures documented for publication and review.
Omics matrices · raw reads in
Factors · clusters · drivers out
Typical turnaround 3–7 days
Individual pricing from $149
Illustrative sample output
variance explained
latent factors × views
Overview
Rigorous integration, without the in-house pipeline overhead
Multi-omics integration turns several separate data layers—expression, methylation, chromatin, protein, metabolite—into one model that captures how they move together. The hard part is rarely running a single tool; it is choosing an integration method suited to your samples and question, normalizing and harmonizing layers that live on completely different scales, and documenting every decision so the result survives peer review.
We build and run that workflow for you. Whether you have matched multi-omics on the same samples or datasets that never shared a sample at all, we harmonize the layers and apply established, peer-reviewed methods—MOFA+, DIABLO, SNF, and single-cell multimodal approaches, never opaque in-house black boxes—and return an interpretable, integrated result with the tool versions and parameters recorded for full reproducibility.
Capabilities
What we analyse
One service spanning the ways omics layers can be combined—factor models, supervised integration, network fusion, and single-cell multimodal.
Supervised biomarker integration
Multiblock sPLS-DA that selects the cross-omics features separating known groups—built for biomarker panels and prediction.
Similarity network fusion
Per-layer sample-similarity networks fused into one—robust integration that works even when layers don't share samples.
Integrative clustering & subtyping
Joint latent-variable clustering that defines molecular subtypes from several data types at once—a mainstay of disease stratification.
Single-cell multimodal
Paired 10x Multiome (RNA + ATAC) and CITE-seq (RNA + protein) integrated into a joint embedding with cell-type resolution.
Regulatory & QTL integration
Linking genotype to expression, methylation, or chromatin—eQTL, mQTL, and caQTL mapping to connect variants to molecular effects.
Network & pathway integration
Co-expression and regulatory networks that place cross-omics features into modules and pathways for mechanistic interpretation.
Spatial & unmatched integration
Diagonal and spatiotemporal integration that aligns modalities without shared samples or across tissue space and time.
The Pipeline
How an integration analysis runs
A transparent, best-practice sequence—each step chosen for your data and question and documented in the final report. Nothing is a black box.
Steps are adapted to your design: matched vs. unmatched samples, two layers vs. five, bulk vs. single-cell, discovery vs. prediction. We confirm the plan with you before any compute begins.
Scope & study design
We define the biological question, map which layers and samples you have, and select an integration strategy to match.
Choices: vertical vs. diagonal · supervised vs. unsupervised
Per-omics QC & preprocessing
Each layer is quality-controlled, normalized on its own terms, and reduced to informative features before anything is combined.
Tools: DESeq2 · limma · per-omics QC
Harmonization & alignment
Samples and features are matched across layers, scales reconciled, and batch effects addressed so layers are comparable.
Tools: MultiAssayExperiment · MuData · ComBat
Integration modelling
The chosen model is fit—factor analysis, supervised multiblock, network fusion, or a single-cell multimodal method.
Tools: MOFA+ · DIABLO · SNF · Seurat WNN
Factor & cluster interpretation
We quantify variance explained per layer, rank feature loadings and drivers, and resolve sample clusters or subtypes.
Outputs: factors · loadings · subtypes
Cross-omics annotation
Integrated signals are connected to pathways, networks, and known biology so the model reads as mechanism, not just math.
Tools: clusterProfiler · STRING · Reactome
Reporting & delivery
Integrated tables, cross-omics figures, a harmonized dataset, and reproducible methods with every tool version.
Tools: ComplexHeatmap · ggplot2 · versioned methods manifest
Tools & Technologies
Established, peer-reviewed tools—matched to your data
We select from the field's standard toolkit rather than forcing every dataset through one pipeline. A representative set of what we work with:
Factor & latent-variable
MOFA+
MOFA2
MEFISTO
iClusterPlus
iClusterBayes
MCIA
Supervised integration
mixOmics
DIABLO
sPLS-DA
sGCCA
RGCCA
Network-based fusion
SNF
NEMO
ANF
netZoo
WGCNA
Single-cell multimodal
Seurat WNN
Signac
muon
LIGER
UINMF
bindSC
Deep generative
totalVI
MultiVI
scVI
GLUE
SpatialGlue
Cobolt
QTL & regulatory
MatrixEQTL
tensorQTL
PANDA
LIONESS
Data containers & handling
MultiAssayExperiment
MuData / AnnData
SummarizedExperiment
Bioconductor
Pathway & visualization
clusterProfiler
STRING
Cytoscape
ComplexHeatmap
UpSetR
Reference Resources
Public knowledge behind every result
An integrated model is only as useful as the biology it connects to. We build on the community's authoritative, versioned resources.
- TCGA & ICGC
- Large matched multi-omics cancer cohorts for reference and benchmarking.
- GTEx
- Tissue-level expression and eQTL reference across the human body.
- CPTAC
- Proteogenomic datasets linking genome, transcriptome, and proteome.
- Ensembl / GENCODE
- Shared gene and feature models that align identifiers across layers.
- GO, KEGG & Reactome
- Ontologies and pathways for interpreting integrated signals.
- MSigDB
- Curated gene-set collections for enrichment of factors and modules.
- STRING
- Protein–protein interaction networks for cross-omics module context.
- GEO / SRA / ArrayExpress
- Public archives for assembling and validating multi-omics datasets.
- PRIDE & MetaboLights
- Reference repositories for proteomics and metabolomics data.
Choosing a Method
MOFA+ vs. DIABLO vs. SNF
There is no single best method—only the right one for your samples and question. A quick orientation; we will help you decide.
| Dimension | MOFA+ | DIABLO | SNF |
|---|---|---|---|
| Approach | Unsupervised Bayesian factor analysis | Supervised multiblock sPLS-DA | Network fusion of similarity graphs |
| Uses labels? | No — finds variation without an outcome | Yes — separates known groups | No — clusters fused samples |
| Sample overlap | Same samples across layers | Same samples across layers | Flexible; tolerant of partial overlap |
| Typical output | Latent factors, variance explained, loadings | Selected biomarker panels, components | Fused network, sample clusters |
| Best suited to | Discovery & unbiased structure | Prediction & biomarker selection | Subtyping across heterogeneous layers |
What You Receive
A complete, documented deliverable
Not just a merged spreadsheet dropped in a folder—every output you need to interpret, publish, and reproduce the work.
- Harmonized multi-omics dataset (MultiAssayExperiment / MuData)
- Integrated model with latent factors or components
- Variance explained per layer and per factor
- Feature loadings & ranked cross-omics drivers (TSV / XLSX)
- Sample clusters / molecular subtypes
- Cross-omics figures (factor heatmaps, networks, loadings)
- Publication-ready methods text with tool versions
Built for reproducibility, not just a result
Every integration follows documented best practices—method matched to your design, each layer normalized on its own terms, and batch and scale effects handled before anything is combined—so shared signal is real, not an artifact of merging. We check that factors and clusters are stable rather than fit to noise. Each run records its tool versions, parameters, and reference builds.
The practical payoff: your methods section writes itself, a reviewer can re-run the analysis, and a result from today can be reproduced a year from now. We will also tell you honestly when a design or sample size won't support the conclusion you're after.
FAQ
Multi-omics integration questions
What researchers and project leads most often ask before starting an integration project.
It combines two or more data types — such as genomics, transcriptomics, epigenomics, proteomics, or metabolomics — into a single analysis. You need it when no single layer explains your biology and you want the coordinated signal across layers rather than a stack of separate results.
It depends on the method. Vertical integration (MOFA+, DIABLO) works best with the same samples measured on multiple layers. Network and diagonal methods (SNF, GLUE) can integrate datasets that don't share samples. We pick the approach to match what you have.
MOFA+ is unsupervised and finds shared factors of variation without using labels. DIABLO is supervised and selects features that separate known groups. SNF fuses sample-similarity networks across layers. We choose based on whether you have an outcome to predict and how your samples overlap.
Yes. We handle paired assays such as 10x Multiome (RNA + ATAC) and CITE-seq (RNA + protein) with Seurat WNN, Signac, muon, and deep-generative models like totalVI and MultiVI, and can align unmatched modalities where needed.
Processed per-omics matrices (expression, methylation, peak, protein, or metabolite tables) with a shared sample sheet — or the raw reads, and we run the upstream single-omics pipelines first. Either way, we return a harmonized, documented integrated dataset.
An integrated model with factor or component tables, variance explained per layer, feature loadings and drivers, sample clusters or subtypes, cross-omics figures, and reproducible methods with every tool version recorded.
Related
The layers we bring together
Have data across several omics layers?
Tell us your organism, data type, and question—we'll scope it honestly, including if a different design would serve you better.