Droplet Microfluidics: Poisson Cell Loading, Singlets and Doublets

A translucent PDMS chip with a flow-focusing junction: droplets form, carry a random number of cells and flow out. Blue empty, green one cell, red two or more. Drag to orbit.

Switch to 2D
Controls

Presets

Parameters

Quality

Camera

Model checks

The science

Why droplets contain a random number of cells. In a flow-focusing chip the aqueous suspension is pinched off by oil into droplets of volume V = (π/6)d³. Cells are independent and sparse, so the number of cells captured by one droplet is Poisson distributed with mean λ = c V (cell concentration c): P(k) = e-λλk/k!. A 70 µm droplet holds about 180 pL, so 106 cells per mL give λ ≈ 0.18.

The trade-off. Empty droplets e-λ are wasted reagent, multiplets 1 - e-λ(1 + λ) give corrupted single-cell data (two cells share a barcode). The number of useful single-cell droplets, λ e-λ, peaks at λ = 1 with 36.8 %, but by then 26 % of droplets are multiplets, so single-cell sequencing kits run at λ ≈ 0.05-0.2 and accept that most droplets are empty: among occupied droplets the multiplet share is about λ/2.

Two views of the same numbers. Counting droplets, a fraction 1 - e-λ (1 + λ) are multiplets; counting cells, a fraction 1 - e-λ of all cells share a droplet (size-biased: crowded droplets hold more cells). Both are checked against the simulation.

Simplifications. Perfectly random, independent capture: no cell clumps or sedimentation in the syringe (which make real loading non-Poisson), no deformation or size distribution of droplets, constant concentration, and an ideal sampler. Beads for barcoding (Drop-seq style) also load as Poisson and double the issue. Checks: normalisation, mean and variance of the law; two independent unit-conversion routes; a 200 000-droplet sampler against the pmf bin by bin; golden-section search for the singlet maximum; the droplet-centred and the cell-centred multiplet fractions.

Loading model...
Loading Blender model...
-
-
Made with 🟠 Blender · Three.js