Single-Cell RNA-seq Analysis for Disease Mechanism Discovery

Bulk RNA sequencing has been a workhorse of transcriptomics for over a decade. It measures gene expression across thousands or millions of cells simultaneously, averaging their signals into a single readout per sample. That averaging is the method's core limitation: a tissue is not a single cell type, and the average obscures the heterogeneity underneath. A small subpopulation driving disease can be drowned out entirely.

Single-cell RNA sequencing (scRNA-seq) resolves this. By capturing and sequencing the transcriptome of individual cells, it reveals cell-type composition, transcriptional states, and gene expression patterns that bulk methods cannot see. Over the last decade, scRNA-seq has shifted from a technically demanding novelty to a broadly deployed tool for mapping disease biology at cellular resolution.

This post covers what scRNA-seq actually measures, how the analysis pipeline works, and what it has revealed about specific disease mechanisms, with an honest account of where the technology's limits currently sit.

What scRNA-seq Measures

Every cell in a tissue carries the same genome. What distinguishes a neuron from a macrophage is not the DNA sequence but which genes are being actively transcribed: the transcriptome. scRNA-seq captures messenger RNA molecules from individual cells, converts them to complementary DNA, sequences them, and produces a count of transcripts per gene per cell.

The dominant commercial platforms use microfluidics to encapsulate individual cells in droplets alongside barcoded beads. Each bead carries a unique cell barcode and molecular identifiers for individual transcripts. After sequencing, these barcodes allow every read to be assigned to the cell it came from. The 10x Genomics Chromium platform is the most widely used implementation and can process thousands of cells in a single run [1].

The resulting data is a cell-by-gene matrix: rows are cells, columns are genes, and each entry is a transcript count. A typical experiment produces data for 5,000 to 50,000 cells, with counts for 20,000 to 30,000 genes per cell, though in practice most genes are not detected in any given cell, making the matrix very sparse.

The Analysis Pipeline

Raw count matrices go through several steps before biological interpretation becomes possible.

Quality control and filtering removes low-quality cells: those with too few detected genes (likely empty droplets), too many (likely two cells captured together, called doublets), or a high proportion of mitochondrial transcripts (often a sign of cell damage). Typical thresholds are set empirically per dataset and can meaningfully affect downstream results [2].

Normalization corrects for the fact that cells with more total transcripts will appear to express every gene more highly. The most common approach scales each cell to a fixed total count and log-transforms the data [3].

Dimensionality reduction compresses the high-dimensional gene expression space into two or three dimensions for visualization and clustering. Principal component analysis (PCA) captures the major axes of variation; UMAP (Uniform Manifold Approximation and Projection) then projects the top principal components into a 2D plot where similar cells cluster together [4].

Clustering groups cells by transcriptional similarity using graph-based methods, most commonly the Leiden or Louvain algorithms, applied to the PCA-reduced data. Each cluster is then annotated as a cell type by matching its top expressed genes to known marker genes.

Differential expression within or across clusters identifies genes upregulated in specific conditions, cell types, or disease states. Tools like DESeq2, edgeR, and more recently methods designed for the sparse count distributions of scRNA-seq data (MAST, Wilcoxon rank-sum with correction) are standard [5].

Seurat (R) and Scanpy (Python/AnnData) are the two most widely used end-to-end frameworks, and most published analyses are reproducible within one or the other [6, 7].

What scRNA-seq Has Revealed About Disease

The value of scRNA-seq for disease research lies in its ability to identify which cells are relevant, not just which genes. Several examples illustrate what this has meant in practice.

Alzheimer's disease. A landmark 2019 study by Mathys et al. in Nature profiled 80,660 single-nucleus transcriptomes from human prefrontal cortex across individuals with varying degrees of Alzheimer's pathology. The analysis identified distinct transcriptional changes in excitatory neurons, oligodendrocytes, and astrocytes that were not apparent in bulk data, and found that early and late pathology involved different cellular programs. Importantly, sex-specific differences in glial response were only visible at single-cell resolution [8].

Inflammatory bowel disease. Smillie et al. (2019, Cell) profiled approximately 50,000 cells from the colon of Crohn's disease and ulcerative colitis patients and healthy controls. They identified a subset of inflammatory fibroblasts and characterized specific epithelial cell losses associated with disease, enabling more precise mechanistic hypotheses than prior bulk data had allowed [9].

COVID-19 and lung injury. Multiple groups used scRNA-seq to characterize the cellular response to SARS-CoV-2 infection. A 2021 study published in Nature mapped the cellular landscape of COVID-19 lung tissue, identifying aberrant macrophage states and impaired alveolar epithelial regeneration as features of severe disease, with direct relevance to understanding why some patients develop fibrotic sequelae [10].

Cancer tumor microenvironments. The tumor microenvironment has been one of the most productive areas for scRNA-seq. Studies across cancer types have identified T cell exhaustion states, distinct macrophage polarizations, and cancer-associated fibroblast subtypes that correlate with immunotherapy response. Work from the Regev, Bhatt, and van Galen labs has been particularly influential in acute myeloid leukemia and solid tumors [11].

These are not incremental findings. They reflect genuinely new biology that bulk methods had missed because the relevant cell populations were too rare or too transcriptionally distinct to emerge from averaged data.

Trajectory Analysis and Cell State Transitions

A significant extension of basic clustering is trajectory inference: methods that order cells along a continuous progression to model transitions between states. Tools like Monocle, RNA velocity (which uses the ratio of spliced to unspliced RNA as a proxy for the direction of transcriptional change), and PAGA are used to reconstruct developmental trajectories, differentiation paths, or disease progressions from static snapshots [12, 13].

RNA velocity, described by La Manno et al. in Nature (2018), is particularly notable: by estimating whether a gene's transcript is being actively produced or degraded, it infers the likely future transcriptional state of each cell, turning a static dataset into a directional map [12].

These methods have real interpretive caveats. Trajectory inference makes assumptions about the underlying biology that may not hold in all tissues. Used carefully, however, they have helped identify transition states relevant to cancer progression, immune activation, and developmental disorders.

Current Limitations

scRNA-seq is powerful but not without constraints that matter for disease biology interpretation.

Dropout. Because each cell is sequenced from a small amount of RNA, many transcripts are not captured at all, producing zero counts for genes that are actually expressed. This technical noise complicates differential expression and trajectory analyses. Imputation methods exist but introduce their own assumptions [2].

Spatial information is lost. scRNA-seq requires cell dissociation, which destroys the spatial relationships between cells. A macrophage from the center of a tumor and one from its periphery are indistinguishable once both are in suspension. Spatial transcriptomics methods (10x Visium, Slide-seq, MERFISH) address this by measuring gene expression in tissue sections, but currently at lower gene detection sensitivity or cell resolution than droplet-based scRNA-seq [14].

Snapshot limitation. A single experiment captures gene expression at one moment. Longitudinal changes in how cells evolve during treatment, remission, or disease progression require multiple timepoints and careful study design.

Batch effects. Samples processed in different laboratories, on different days, or with different protocols show systematic differences in their data that can mimic biological signal. Batch correction tools (Harmony, BBKNN, scVI) are standard but imperfect [15].

Cell type annotation is not standardized. Different groups annotate the same clusters differently, making cross-study comparison difficult. Community efforts like the Human Cell Atlas and CellTypist are working to establish reference atlases and automated annotation, but the field is not yet there for most tissue types [16].

Connecting scRNA-seq to Drug Discovery

The relevance to drug development lies in a cleaner answer to a central question: which cell type, in which state, is the right target? Identifying a disease-relevant transcriptional program in a specific subpopulation, rather than a bulk tissue average, allows target identification to be more precise. Combining scRNA-seq with genetic data (eQTL mapping, GWAS colocalization) to ask whether a candidate gene's expression in the relevant cell type is genetically linked to disease risk has become a standard early-stage validation step for new targets.

The technology has moved from generating insight to generating testable hypotheses at scale. That shift is what makes it a genuine tool for disease mechanism discovery, rather than a sophisticated descriptive exercise.

---

References

1. Zheng GX et al. Massively parallel digital transcriptional profiling of single cells. Nature Communications. 2017;8:14049. https://doi.org/10.1038/ncomms14049
2. Luecken MD, Theis FJ. Current best practices in single-cell RNA-seq analysis: a tutorial. Molecular Systems Biology. 2019;15(6):e8746. https://doi.org/10.15252/msb.20188746
3. Hafemeister C, Satija R. Normalization and variance stabilization of single-cell RNA-seq data using regularized negative binomial regression. Genome Biology. 2019;20:296. https://doi.org/10.1186/s13059-019-1874-1
4. McInnes L, Healy J, Melville J. UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction. arXiv. 2018. https://arxiv.org/abs/1802.03426
5. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biology. 2014;15:550. https://doi.org/10.1186/s13059-014-0550-8
6. Hao Y et al. Integrated analysis of multimodal single-cell data. Cell. 2021;184(13):3573–3587. https://doi.org/10.1016/j.cell.2021.04.048
7. Wolf FA, Angerer P, Theis FJ. SCANPY: large-scale single-cell gene expression data analysis. Genome Biology. 2018;19:15. https://doi.org/10.1186/s13059-017-1382-0
8. Mathys H et al. Single-cell transcriptomic analysis of Alzheimer's disease. Nature. 2019;570:332–337. https://doi.org/10.1038/s41586-019-1195-2
9. Smillie CS et al. Intra- and inter-cellular rewiring of the human colon during ulcerative colitis. Cell. 2019;178(3):714–730. https://doi.org/10.1016/j.cell.2019.06.029
10. Melms JC et al. A molecular single-cell lung atlas of lethal COVID-19. Nature. 2021;595:114–119. https://doi.org/10.1038/s41586-021-03569-1
11. van Galen P et al. Single-cell RNA-seq reveals AML hierarchies relevant to disease progression and immunity. Cell. 2019;176(6):1265–1281. https://doi.org/10.1016/j.cell.2019.01.031
12. La Manno G et al. RNA velocity of single cells. Nature. 2018;560:494–498. https://doi.org/10.1038/s41586-018-0414-6
13. Trapnell C et al. The dynamics and regulators of cell fate decisions are revealed by pseudotemporal ordering of single cells. Nature Biotechnology. 2014;32:381–386. https://doi.org/10.1038/nbt.2859
14. Moses L, Pachter L. Museum of spatial transcriptomics. Nature Methods. 2022;19:534–546. https://doi.org/10.1038/s41592-022-01409-2
15. Korsunsky I et al. Fast, sensitive and accurate integration of single-cell data with Harmony. Nature Methods. 2019;16:1289–1296. https://doi.org/10.1038/s41592-019-0619-0
16. Domínguez Conde C et al. Cross-tissue immune cell analysis reveals tissue-specific features in humans. Science. 2022;376(6594):eabl5197. https://doi.org/10.1126/science.abl5197