By EVOBYTE Your partner in bioinformatics
Introduction
If bulk genomics is like listening to a crowd, single‑cell omics is walking through the room and asking each person what they think—and where they’re standing. That shift in perspective has changed how we study development, immunity, and disease. Yet “single‑cell omics” can feel like a maze of acronyms and fast‑moving tools. In this guide, we’ll demystify the core technologies—scRNA‑seq, scATAC‑seq, CITE‑seq, Multiome, and spatial transcriptomics—and show how modern analysis frameworks bring them together into one coherent view of biology. Along the way, we’ll ground ideas with short code snippets you can try as a starting point and a concrete example you can picture.
Imagine a tumor biopsy. With single‑cell omics, you can profile the gene programs that drive each cell’s behavior, peek into the regulatory DNA that switches those programs on and off, measure key surface proteins that shape immune interactions, and then place every measurement back into its original tissue neighborhood. That’s the power of single‑cell multi‑omics—and it’s increasingly within reach.
Reading the messages and the map: scRNA‑seq and scATAC‑seq
Single‑cell RNA‑seq (scRNA‑seq) measures gene expression in individual cells, giving you a “what’s happening now” snapshot of cell state. It’s the workhorse for discovering cell types, inferring trajectories, and parsing heterogeneous tissues.
To understand why cells choose certain fates, you also want to see the regulatory landscape. Single‑cell ATAC‑seq (scATAC‑seq) profiles open chromatin, highlighting where transcription factors can bind and which enhancers may be active. Together, scRNA‑seq and scATAC‑seq connect cellular identity to the underlying control logic—messages and map. Platforms now support joint profiling from the same nucleus, aligning expression and accessibility one‑to‑one, which strengthens regulatory inference and simplifies integration later.
Consider that tumor biopsy again. scRNA‑seq might flag a cluster of exhausted CD8 T cells; scATAC‑seq from the same cells can surface motif activity for regulators like TOX or NFAT, supporting a mechanistic story rather than just a label. That combination often turns descriptive atlases into actionable hypotheses.
Beyond transcripts: CITE‑seq and true multimodal single‑cell data
mRNA tells part of the story, but cell‑surface proteins are how cells sense and signal in real time. CITE‑seq (Cellular Indexing of Transcriptomes and Epitopes by sequencing) attaches DNA barcodes to antibodies, allowing simultaneous measurement of transcripts and surface proteins per cell. It bridges the gap between high‑dimensional sequencing and immunology’s familiar protein markers, improving cell type resolution and making it easier to align results with flow cytometry or pathology panels.
Meanwhile, “Multiome” assays jointly capture chromatin accessibility (scATAC‑seq) and gene expression (scRNA‑seq) from the same nucleus. Because the modalities come from the same cell, you can directly link an open enhancer to a nearby gene’s expression without guesswork. In practice, that means clearer regulatory networks in development, immunity, and cancer—fewer ambiguous mappings and more convincing causal stories.
Putting cells back on the slide: spatial transcriptomics at single‑cell scale
Dissociated single‑cell data breaks cells out of tissue, which helps with resolution but loses context. Spatial transcriptomics brings that context back by measuring gene expression with coordinates in the original tissue section. Newer platforms now reach single‑cell‑scale resolution on formalin‑fixed (FFPE) samples, making it easier to connect morphology, microenvironment, and molecular programs. Researchers are using these maps to trace tumor–immune niches, layer‑specific neuronal programs, and gradients in inflamed tissues.
Returning to our biopsy, spatial data lets you ask where those exhausted T cells sit: hugging tumor borders, grouped around vasculature, or interleaved with myeloid cells. When you overlay spatial patterns with multimodal single‑cell labels, you can pinpoint microenvironments that matter for therapy response and resistance.
From silos to synthesis: integrating scRNA‑seq, scATAC‑seq, CITE‑seq, Multiome, and spatial data
Great measurements only pay off if we can integrate them cleanly. Two practical strategies dominate today’s toolkits.
First, graph‑based multimodal integration learns how much each modality should contribute per cell. Weighted Nearest Neighbor (WNN) analysis, popularized in Seurat, builds per‑modality neighbor graphs and learns cell‑wise weights that fuse them into a single representation. The result is a joint manifold that respects RNA, protein, and chromatin signals without letting any one drown out the others. It’s simple, fast, and widely adopted for CITE‑seq and Multiome datasets.
Second, probabilistic deep learning models estimate a shared latent space while explicitly accounting for noise, sparsity, and batch effects. The scvi‑tools ecosystem offers off‑the‑shelf models for scRNA‑seq, scATAC‑seq, CITE‑seq, Multiome, and spatial data, with GPU acceleration and clean APIs that plug into AnnData/Scanpy. These models help with integration, differential testing, annotation, and generative tasks, all within a principled Bayesian framework.
Here’s a minimal, end‑to‑end sketch in Python showing how you might build an integrated latent space with scvi‑tools:
import scanpy as sc
import scvi
adata = sc.read_h5ad("tumor_multiome.h5ad") # RNA + ATAC in one AnnData with layers/obsm
scvi.model.MULTIVI.setup_anndata(
adata, batch_key="batch", rna_layer="counts",
atac_layer="peaks", modality_key="modality"
)
model = scvi.model.MULTIVI(adata)
model.train(max_epochs=100)
adata.obsm["X_latent"] = model.get_latent_representation()
sc.pp.neighbors(adata, use_rep="X_latent"); sc.tl.umap(adata)
And if you prefer the WNN route in Seurat for a CITE‑seq or RNA+ATAC object:
library(Seurat)
# Assume object has RNA (PCA dims) and ATAC (LSI dims) reductions
object <- FindMultiModalNeighbors(
object, reduction.list = list("pca", "lsi"), dims.list = list(1:30, 2:30)
)
object <- RunUMAP(object, nn.name = "weighted.nn", reduction.name = "wnn.umap")
These approaches aren’t mutually exclusive. Many teams explore both—using WNN for quick iteration and visualization, and a probabilistic model to sharpen integration, quantify uncertainty, and improve downstream tests.
Finally, large reference atlases are emerging that blend modalities and developmental stages. Mapping your data into these references can speed annotation and highlight rare states you might otherwise miss. Think of it as standing on the shoulders of a curated, multi‑omics compendium rather than rebuilding from scratch.
A quick story to make it stick
A clinician‑scientist profiles a lung tumor before immunotherapy. scRNA‑seq reveals exhausted CD8 T cells and an inflamed myeloid compartment. CITE‑seq shows low PD‑1 protein on many T cells despite high PDCD1 mRNA, hinting at post‑transcriptional regulation and suggesting that a PD‑L1–only strategy might underperform. Multiome links an accessible enhancer near CXCL13 to a tertiary lymphoid‑like T cell subset, while spatial maps locate these cells adjacent to B‑cell aggregates at the invasive margin. Integrated together with WNN and a variational model, the picture supports a combination strategy that targets myeloid checkpoints and leverages tertiary lymphoid niches. Each modality contributes a piece; integration turns pieces into a plan.
Summary / Takeaways
Single‑cell omics is no longer just scRNA‑seq. When you pair expression with chromatin, proteins, and spatial context—and then integrate with modern methods—you move from snapshots to mechanisms. Start with a clear question, pick the modalities that best answer it, and plan analysis early so integration feels routine, not heroic. If you’re new to multimodal analysis, try WNN for quick wins and scvi‑tools for robust modeling. Then, map into a public atlas to benchmark your findings and spot rare biology with confidence.
What tissue or question are you tackling first? If you share a brief description, I can suggest a minimal multimodal design and an analysis path tailored to your goal.
Further Reading
- Integrated analysis of multimodal single‑cell data (WNN in Seurat), Cell (2021). https://www.sciencedirect.com/science/article/pii/S0092867421005833
- Single Cell Atlas: a single‑cell multi‑omics human cell encyclopedia, Genome Biology (2024). https://genomebiology.biomedcentral.com/articles/10.1186/s13059-024-03246-2
- CITE‑seq: simultaneous epitope and transcriptome measurement, Nature Methods (2017). https://pmc.ncbi.nlm.nih.gov/articles/PMC5669064/
- scvi‑tools: probabilistic modeling for single‑cell omics (docs and tutorials). https://scvi-tools.org/get_started/
- Scoping review of spatial transcriptomics methods and applications in cancer, Cancers (2024). https://www.mdpi.com/2072-6694/16/17/3100