Loading field map
For most of genomics' history, the fundamental unit of analysis was a population of cells—a tissue biopsy, a blood draw, a culture flask. The resulting measurements were averages across thousands or millions of cells, obscuring the differences that make each cell unique. A tumor sample might appear to have intermediate expression of a gene when, in reality, half the cells express it strongly and the other half not at all. This averaging problem drove the development of a new subfield: single cell genomics, which sought to measure genomes, transcriptomes, and other molecular features one cell at a time. Over two decades, five successive research paradigms have redefined what it means to study individual cells, each building on the technical and conceptual foundations of its predecessors while opening questions the earlier approaches could not address.
The first paradigm, the Single-Cell Genome Amplification Paradigm (roughly 2000–2015), tackled a deceptively simple question: can we sequence the DNA of a single cell? The obstacle was that a single human cell contains only about 6 picograms of DNA, far too little for conventional sequencing. The solution was whole-genome amplification (WGA), and the most influential technique was multiple displacement amplification (MDA), which used the phi29 polymerase to copy the genome isothermally, producing fragments up to 100 kilobases long. MDA made it possible to sequence individual cells for the first time, opening applications in cancer genomics (where researchers could trace tumor evolution cell by cell), preimplantation genetic diagnosis, and the study of unculturable microbes. Yet the paradigm also revealed a persistent limitation: amplification bias. Not all regions of the genome amplified evenly, leading to uneven coverage and missing alleles. This bias narrowed the paradigm's scope—it remained useful for detecting large copy-number variants and for microbial genomics, but it could not deliver the uniform, high-coverage genomes needed for single-nucleotide variant detection. The Amplification Paradigm was not so much replaced as absorbed into a narrower niche; its infrastructure, particularly MDA, continued to serve as a preparatory step for later single-cell DNA sequencing methods, but the field's center of gravity shifted decisively toward RNA.
The Single-Cell Transcriptomics Paradigm (2009–present) emerged from a different pressure: the need to understand cell states, not just cell genomes. RNA is far more abundant than DNA in a cell, and reverse transcription could convert it into amplifiable cDNA. The first single-cell RNA sequencing (scRNA-seq) methods, such as Smart-seq and CEL-seq, used plate-based protocols that captured full-length transcripts but could process only dozens or hundreds of cells per experiment. A fierce debate soon arose between depth and scale. One camp argued that deep coverage per cell was essential to detect lowly expressed genes and splice variants; the other camp insisted that capturing thousands of cells was necessary to discover rare cell types and to map developmental trajectories. The scale camp won a decisive victory with the introduction of droplet-based methods, most notably Drop-seq and the commercial 10x Genomics platform, which encapsulated single cells in nanoliter droplets with barcoded beads. These methods could profile tens of thousands of cells in a single run, albeit with lower sensitivity per cell. The paradigm's crowning achievements were large-scale cell atlas projects—the Human Cell Atlas, the Mouse Cell Atlas—that cataloged cell types across entire organisms. But the Transcriptomics Paradigm also had a built-in limitation: it measured only one molecular layer. A cell's identity and behavior are shaped by its genome, epigenome, proteome, and spatial context, none of which scRNA-seq captured. This single-layer constraint did not replace the paradigm—scRNA-seq remains the most widely used single-cell method today—but it motivated a push toward broader molecular coverage.
The Single-Cell Multi-omics Paradigm (2015–present) asked: can we measure multiple molecular layers from the same cell? The pressure was that a cell's transcriptome alone could not distinguish between, say, a regulatory change driven by DNA methylation and one driven by chromatin accessibility. Early multi-omic methods, such as scNMT-seq (which simultaneously profiled DNA methylation, chromatin accessibility, and the transcriptome), showed that it was technically possible but came with a steep tradeoff: each additional layer reduced coverage quality for every other layer. A different strategy, represented by CITE-seq, used oligonucleotide-conjugated antibodies to measure surface proteins alongside the transcriptome, adding a proteomic layer without sacrificing RNA coverage. The Multi-omics Paradigm redefined cell state as a composite of multiple molecular dimensions, but it also introduced a new tension: the more layers one measures, the fewer cells one can process, and the noisier each measurement becomes. This paradigm did not replace the Transcriptomics Paradigm; rather, it coexists with it, offering deeper molecular resolution at the cost of throughput. Researchers now choose between single-layer scale and multi-layer depth depending on their question. The paradigm also created a bridge to the next framework by highlighting that even multi-omic measurements from dissociated cells miss a critical dimension: where each cell sits in the tissue.
The Spatial Single-Cell Genomics Paradigm (2016–present) grew from the recognition that dissociation destroys spatial context. A T cell interacting with a tumor cell, a neuron receiving synaptic input from a specific partner—these relationships are invisible once cells are in suspension. Spatial methods split into two broad families. Imaging-based approaches, such as MERFISH and seqFISH, used sequential fluorescence hybridization to localize hundreds to thousands of RNA species at subcellular resolution. Sequencing-based approaches, such as Slide-seq and Visium, placed tissue sections on barcoded arrays that captured RNA from defined spatial coordinates, trading resolution for transcriptome-wide coverage. The central tension in this paradigm is between resolution and throughput: imaging methods achieve single-cell or even subcellular resolution but are limited in gene number, while sequencing methods capture the whole transcriptome but at lower spatial resolution. The Spatial Paradigm partially overlaps with the Multi-omics Paradigm—both aim for a more complete picture of cell state—but they differ in where measurements happen: multi-omics adds molecular layers in dissociated cells, while spatial genomics adds a tissue coordinate. The two are increasingly converging: spatial multi-omic methods now measure transcripts and proteins, or transcripts and chromatin accessibility, on the same tissue section. This convergence is reshaping tissue biology, enabling researchers to map cell-cell interactions, tumor microenvironments, and neural circuits with molecular precision.
The Single-Cell Functional Genomics Paradigm (2018–present) shifted the goal from measuring cells to manipulating them. Observational single-cell methods could describe cell states, but they could not establish causality: does a transcription factor drive a cell's identity, or is it merely correlated with it? The paradigm's signature approach, Perturb-seq, combined CRISPR-based gene perturbation with scRNA-seq readout, delivering a guide RNA and a barcode into each cell so that the transcriptomic effect of a knockout could be measured. This framework absorbed the infrastructure of the Transcriptomics Paradigm—the readout was essentially scRNA-seq—but added a causal intervention layer. The central challenge became scale: perturbing thousands of genes across millions of cells required combinatorial barcoding and computational methods to deconvolve mixed perturbations. The Functional Paradigm also began to converge with the Spatial Paradigm: spatial perturbation screens now knock out genes in defined tissue regions and read out the effects in situ, blurring the boundary between functional and spatial genomics. This paradigm remains the youngest and most rapidly evolving, driven by the ambition to build a causal map of gene function in every cell type.
Today, four of the five paradigms remain active, each with a distinct research program. The Transcriptomics Paradigm dominates in scale: it is the workhorse for cell atlas projects, disease profiling, and developmental biology, where capturing tens of thousands of cells is essential. The Multi-omics Paradigm leads in depth: it is the method of choice for understanding regulatory mechanisms, where measuring DNA methylation, chromatin accessibility, and RNA from the same cell reveals causal chains. The Spatial Paradigm excels in context: it is indispensable for tissue architecture, tumor immunology, and neuroscience, where cell position determines function. The Functional Paradigm drives causation: it is the primary tool for validating gene function and mapping regulatory networks. The Amplification Paradigm persists in a narrow but important niche: single-cell DNA sequencing for cancer evolution and microbial genomics.
These paradigms agree on a core principle: cellular heterogeneity is not noise but signal, and measuring it requires methods that preserve individual cell identity. They disagree on what constitutes a complete measurement. The Transcriptomics Paradigm treats the transcriptome as sufficient for defining cell types; the Multi-omics Paradigm insists that multiple molecular layers are necessary; the Spatial Paradigm argues that context is as important as content; and the Functional Paradigm holds that observation without perturbation is incomplete. These disagreements are productive: they drive the development of hybrid methods that combine scale, depth, space, and function. A researcher today might use scRNA-seq for discovery, spatial transcriptomics for localization, and Perturb-seq for validation—all within the same study. The boundaries between paradigms are blurring, and the next generation of single-cell genomics will likely integrate all four active frameworks into unified experimental designs.