Skip to Content

 



Single-Cell Multi-Omics Integration Using Bioinformatics Pipelines 


Introduction

Biological systems are inherently complex, resulting from interactions across multiple molecular levels such as the genome, epigenome, transcriptome, proteome, and metabolome. Traditional omics approaches, based on mass analysis, average the signals from millions of cells, thus masking cellular heterogeneity and obscuring rare or transient cellular states.

The emergence of single-cell technologies has transformed the life sciences by enabling molecular profiling at the cellular level. However, each single-cell modality provides only a partial view of cellular identity. 

Single-cell multi-omics integration, supported byAdvanced bioinformatics pipelines address this limitation by combining multiple layers of data to build a unified, high-resolution representation of cellular systems.

 

Biomolecules in various types of biological samples and modalities of data captured by single-cell multi-omics in biomedical (e.g., disease progression and hematopoiesis) and environmental (e.g., microbial responses to environmental stimuli) studies.

What is single-cell multi-omics analysis?

The single-cell multi-omics approach refers to the simultaneous or coordinated measurement of several molecular modalities from individual cells. These modalities can originate from the same cell (true multi-omics approach) or from matched cell populations.

Common single-cell omics layers include:

  • scRNA-seq: gene expression profiling
  • scATAC-seq:chromatin accessibility
  • scDNA-seq: genomic variation and copy number changes
  • scProteomics / CITE-seq: surface and intracellular protein levels
  • Spatial transcriptomics : spatial context of gene expression
  • Single-cell methylomics:DNA methylation profiles

Each modality provides complementary biological information. Integration allows researchers to link regulatory mechanisms to functional outcomes , rather than analyzing them in isolation.


Data profile obtained for each basic method. The profile obtained for each single-cell profiling method reflects that of the associated basic method. The figure format is inspired by that of the article (Ja-Rang Lee et al., 2018) [45]. (a) Example of the final data profile obtained with the single-cell method based on RRBS. In accordance with the basic principle of RRBS, reads are mainly observed at CG-rich positions. (b) Example of the final data profile obtained with the single-cell method based on WGBS. In accordance with the basic principle of WGBS, a relatively homogeneous distribution of reads is observed. (c) Example of the results profile obtained by (sc)CGI-seq with methylation-sensitive restriction enzymes (MSREs). As with RRBS, reads are observed at CG-rich positions in the genome, but methylated sites are identified by comparison between the control and the sample.
 Single-cell methylomics

scRNA-seq
Spatial transcriptomics is a method for visualizing and quantitatively analyzing the transcriptome in individual tissue sections. By depositing histological sections onto glass slides coated with oligonucleotides containing positional barcodes, it is possible to generate high-quality cDNA libraries with precise positional information for RNA sequencing. This provides transcriptomic data in a versatile format for bioinformatic analyses of gene expression within tissue, which will be invaluable for both research and diagnostics. Spatial transcriptomics is the first method to provide a comprehensive and unbiased transcriptome analysis, integrating spatial information from tissue sections.
 spatial transcriptomics


The input data for MaCroDNA consists of tables of scRNA-seq gene expression read counts (or their logarithmically transformed values) and scDNA-seq absolute copy numbers (or their logarithmically transformed values), assumed to be obtained from the same tissue. Distinct clones are distinguished by their color saturation. From the gene expression and copy number matrices (pixel color intensity is proportional to the copy number/gene expression ratio), MaCroDNA identifies the association of scRNA-seq cells (dark orange circles) with gene expression values ​​to scDNA-seq cells (dark blue circles) with copy number values. When the number of scRNA-seq cells exceeds the number of scDNA-seq cells (as in the previous example), MaCroDNA infers the correspondence between cells (indicated by pink arrows) by solving a series of maximal weighted bipartite matching problems (in this example, two steps are required). First, only six scRNA-seq cells (the same number as scDNA-seq cells) are assigned to the scDNA-seq cells in such a way that no pair of scRNA-seq cells is associated with the same scDNA-seq cell, and the sum of the Pearson correlation coefficients between the pairs is maximized. The scRNA-seq cells whose matches were identified in the previous step are removed. Then, the three remaining scRNA-seq cells are assigned to the best scDNA-seq cells one at a time, according to the same correlation-based criterion. d Clones are inferred from scDNA-seq data using a chosen algorithm (clones are represented by blue, orange, and green bubbles). e Knowing the intercellular correspondences and clonal assignment of scDNA-seq cells, we can assign scRNA-seq cells to scDNA-seq clones. The clonal assignment of an scRNA-seq cell is that of its corresponding scDNA-seq cell. This figure was created with dscDNA-seq :
Illustration of DNA barcode antibodies used in CITE-seq. (B) Schematic representation of CITE-seq combined with Drop-seq. Cells are incubated with antibodies, washed, and then passed through a microfluidic chip where a cell and a bead are sometimes encapsulated in the same droplet. After cell lysis, antibody mRNAs and oligonucleotides hybridize to oligonucleotides present on Drop-seq beads, thus linking cell barcodes to cell transcripts and antibody-derived oligonucleotides [adapted from Stoeckius et al.]. CITE-seq


ATCA sequencing scATAC-seq
 The video presents Iain Macaulay's lecture on G&T-seq, a single-cell multi-omics technique that physically separates and sequences genomic DNA and complete transcriptomes from a single cell in parallel to correlate copy number variations, fusions, and aneuploidy with gene expression. It illustrates applications in cancer cell lines and mouse embryos, while also exploring extensions to epigenomics, such as methylation and chromatin accessibility, for a deeper understanding of cell dynamics.

Why multi-omics integration is essential

Single-cell data are multidimensional, sparse, and noisy. Although each omics layer is informative, it cannot fully explain cell behavior on its own.

Multi-omics integration enables:

  • Identification of cell states and types with greater confidence
  • Link between regulatory elements (chromatin, methylation) and gene expression
  • Analysis of development trajectories and lineage decisions
  • Resolution of disease-specific cellular programs
  • Improving biological interpretability beyond simple grouping

From the perspective of systems biology, integration moves the analysis from descriptive profiling to mechanistic understanding .


Challenges related to single-cell multi-omics data

Integrating single-cell multi-omics data is computationally demanding due to several inherent challenges:

1. Data heterogeneity

The different omics layers have distinct characteristics:

  • Feature spaces (genes, peaks, proteins)
  • Data distributions
  • Noise and parsimony of measures

Such heterogeneity complicates alignment and joint analysis.                            

                                                  ​


3. Dimensionality and parsimony

Single-cell data typically contains:

  • Thousands of characteristics per cell
  • High school dropout rates
  • Low signal-to-noise ratios

One of the main objectives of integration pipelines is to reduce dimensionality while preserving the informative biological structure.

                                                       

2. Batch effects and technical biases

Datasets are often generated through:

  • Different platforms
  • Separate experiences
  • Variable sequencing depths

Effective integration requires careful correction of batch and technique-related variations to avoid confusion in biological interpretation.

                                                  


4. Scalability

Modern experiments can generate millions of cells , which requires efficient and reproducible processing chains implemented with scalable computing methods.

  


                                                        



----------------------------------------------

Pipelines bioinformatiques

for multi-omics integration


A bioinformatics pipeline is a structured computing workflow that transforms raw sequencing data into biologically relevant results. For single-cell multi-omics analyses, pipelines typically include the following steps:

 Data preprocessing and quality control
















 

Each modality is processed independently before integration. Typical steps include:

  • Alignment of readings and quantification of characteristics
  • Filtering of low-quality cells
  • Removal of duplicates and technical artifacts
  • Standardization and scaling.

Quality indicators differ depending on the modality, but are essential for reliable integration.


Figure: Experimental design and processing of multi-species sc/snRNA-Seq data 

 Selection and representation of features


To reduce noise and dimensionality:

  • Highly variable genes are selected for single-cell RNA sequencing (scRNA-seq).
  • Informative chromatin regions are selected for scATAC sequencing.
  • Protein markers are selected for antibody-based assays. 

                     

Figure: Interpretable and Expressive Latent Representations of Multi-Modal Single-Cell Genomics Data

 
Data are often transformed into latent representations using methods such as principal component analysis (PCA), non-negative matrix factorization (NMF), variational autoencoders, and other machine learning techniques.

 Intermodal alignment

Integration methods aim to align datasets in a shared latent space where biologically similar cells cluster together, regardless of modality.

Common calculation strategies include:

  • Canonical correlation analysis (CCA) and its variants, such as two-order CCA, are used to find shared variations between modalities.
  • The mutual nearest neighbors (MNN) method aligns cells by matching the nearest neighbors in the datasets.

Matrix factorization approaches (e.g., integrative non-negative matrix factorization) identify shared latent factors.

  • Graph-based alignment  establishes graphical relationships between cells across different modalities.
  • Deep learning and variational models use autoencoders and neural networks to learn joint representations.


Figure: Integration of single-cell Multiome datasets


The goal is not to blindly merge data, but to preserve the biological structure while eliminating technical variations .

 Integrated analysis and interpretation


Once aligned, the integrated data can be used to:

  • Cell type annotation
  • Trajectory and pseudotime inference
  • Reconstruction of the regulatory network
  • Differential analysis of states
  • Correlation of intermodal characteristics

Integration enables biological discovery based on hypotheses and data.

  Here is a recent  guide on multi-omic single-cell data joint analysis, covering basic practices and key results.

A starting guide on multi-omic single-cell data joint analysis: basic practices and results

 View ORCID Profile Lorenzo Martini,  View ORCID Profile Roberta Bardini, Stefano Di Carlo 

doi: https://doi.org/10.1101/2024.03.30.587427

This article is a preprint and has not been certified by peer review

Abstract

Multi-omics single-cell data represent an excellent opportunity to investigate biological complexity in general and generate new insights into the biological complexity of heterogeneous multicellular populations. Considering one omics pool at a time captures partial cellular states, while combining data from different omics collections allows for a better reconstruction of the intricacies of cell regulations at a particular time. However, multi-omics data provide only an opportunity. Computational approaches can leverage such opportunities, given that they raise the challenge of consistent data integration and multi-omics analysis. This work showcases a bioinformatic workflow combining existing methods and packages to analyze transcriptomic and epigenomic single-cell data separately and jointly, generating a new, more complete understanding of cellular heterogeneity.

Pipeline frameworks and examples

Several frameworks and computing pipelines have been developed for real-world single-cell multi-omics integration:

  • Seurat with CCA/MNN integrationwidely used to align multiple modalities and suppress batch effects.  

  • Smithan R pipeline that integrates multiple single-cell multi-omics samples while preserving biological signals and efficiently correcting for batch effects.

  • OM-IC (Orthogonal Multimodality Integration and Clustering)  integrates and groups multimodal data with quantitative interpretability.

  • Graph convolution network-based integration frameworks using graph structures to efficiently align omics layers.


These tools illustrate how integration pipelines can combine statistical methods, machine learning, and graphical modeling to address various integration challenges.

Innovative directions in multi-omics integration

The field continues to evolve, with several innovative directions:

Deep learning-based integration

Advanced neural network models and machine learning architectures enable:

  • Nonlinear integration that captures complex intermodal relationships
  • Improved denoising and learning of latent representations
  • Predicting missing categories from partial data

These methods go beyond linear strategies like PCA and CCA to capture more biologically relevant patterns.

Temporal and dynamic modeling

Integration across different moments or dynamic disturbances allows:

  • Signaling dynamics analysis
  • Characterization of decisions relating to cell fate
  • Reconstruction of time regulation programs

This allows us to understand dynamic processes such as differentiation or response to treatment, and not just static snapshots.

Spatial resolution multi-omics

Emerging methods combine molecular profiling with spatial coordinates, enabling:

  • Tissue architecture mapping
  • Elucidating the influences of the microenvironment on the cellular phenotype
  • Deduce cell-cell interactions in a spatial context

These approaches improve biological interpretability by simultaneously integrating spatial and molecular layers.

Applications in research and biology

Single-cell multi-omics integration is a powerful tool for exploring and analyzing cellular systems in various biological contexts:

  • Developmental biology studies the progression of cell lineages and the mechanisms of cell differentiation.
  • Advanced cell biology : analysis of the diversity and organization of cell populations in different experimental environments.
  • Immunology characterizing the variety of cellular states in experimental models of immune responses.
  • Neuroscientific mapping of complex cellular networks and interactions between different cell types.
  • Computational biology and genomics — linking molecular profiles to explore correlations between different data layers and gain a comprehensive understanding of cellular systems.

These integrative approaches allow for a detailed and systemic understanding of cellular behaviors , offering insight into fundamental biological mechanisms, without reference to health outcomes or medical treatments .