The science behind single-cell sequencing

01 MEASURING THE TRANSIENT LAYER

The science behind single-cell sequencing begins with a choice of what to measure. DNA is essentially identical across the cells of an organism and therefore reveals little about current cell state. Proteins carry out function but are difficult to capture and count at scale. RNA occupies the middle of the central dogma: it is the transient readout of which genes a cell is actively expressing at the moment of measurement.

This temporal specificity is both the strength and the vulnerability of single-cell RNA-seq. A cell captured at noon and another at 12:01 may show different transcriptomes not because they are different cell types but because gene expression fluctuates. The science must therefore account for stochasticity as a feature of biology, not merely as noise in the measurement.

02 POLYADENYLATION AS A CAPTURE HANDLE

Messenger RNA in eukaryotes carries a poly(A) tail added during processing. This evolutionary accident becomes an engineering asset: oligo-dT primers can capture polyadenylated RNA without needing to know its sequence in advance. The poly(A) tail is a molecular handle that lets the protocol target the transcriptome without prior design of millions of gene-specific probes.

This convenience introduces bias. Non-polyadenylated transcripts, including many non-coding RNAs and bacterial RNAs, are invisible to oligo-dT capture. Degraded RNA with broken poly(A) tails is lost. The poly(A) handle is therefore a filter, and the science of single-cell sequencing must acknowledge what that filter removes.

03 REVERSE TRANSCRIPTION AND ITS NOISE

Reverse transcriptase converts captured RNA into complementary DNA. This step is imperfect: it has a capture efficiency of roughly 10 to 30 percent in droplet-based protocols, meaning most transcripts in the cell are never measured. The ones that are captured represent a stochastic sample. A cell expressing 10,000 transcripts may yield only 1,500 detected genes, not because the others are absent but because the capture missed them.

This phenomenon is called dropout: a gene that is expressed may appear as zero in a given cell simply because its transcripts were not captured. Dropout is not a bug to be eliminated but a statistical property of the measurement that downstream analysis must model.

04 THE UMI PRINCIPLE

Polymerase chain reaction amplifies captured molecules unevenly. A transcript that receives early amplification may produce hundreds of copies, while a neighboring transcript produces only dozens. Without correction, this would make expression counts reflect amplification bias rather than biological abundance.

The unique molecular identifier solves this. Each original transcript receives a random short barcode before amplification. After sequencing, reads sharing the same UMI are recognized as PCR duplicates of one starting molecule and collapsed to a count of one. The UMI transforms the assay from counting sequenced reads into counting original molecules, a fundamentally more accurate measurement.

05 THE POISSON NATURE OF CAPTURE

The capture of individual transcripts is approximately a Poisson process: each transcript has a small, independent probability of being captured and converted into a measurable molecule. This statistical fact has consequences. Low-abundance transcripts are captured unreliably, producing zeros that do not reflect biological absence. High-abundance transcripts are captured more consistently. There is an effective detection floor below which single-cell measurements become unreliable.

This is why single-cell data is sparse compared to bulk data. The sparsity is not primarily a failure of the instrument but a consequence of sampling statistics applied to finite, small numbers of molecules per cell. Better instruments raise the floor but do not eliminate it.

06 FROM MOLECULES TO MEANING

After sequencing, the bioinformatic pipeline faces a matrix: thousands of cells by tens of thousands of genes, mostly zeros. Dimensionality reduction methods project this high-dimensional data into lower-dimensional spaces where cell types form clusters. Clustering is not a purely computational exercise; it encodes biological hypotheses about what distinguishes cell states from one another.

The science requires careful validation. A cluster might represent a true cell type, a transient state, an artifact of dissociation, or a batch effect. Marker genes, known biology, and orthogonal validation help distinguish signal from structure imposed by the measurement itself.

07 WHAT THE MEASUREMENT CAN AND CANNOT PROVE

Single-cell sequencing is a snapshot. It measures a cell at the moment of lysis and cannot observe the same cell again. Trajectory inference methods attempt to reconstruct developmental sequences from populations of cells frozen at different stages, but these are statistical reconstructions, not direct observations of dynamics.

The science is most powerful when it acknowledges its limits. Single-cell sequencing can reveal heterogeneity with unprecedented resolution, but it cannot by itself establish causation, function, or temporal order. Combining it with perturbation, imaging, and longitudinal sampling is where the biology advances.