Fig 1.
Illustration of multiple fluorescence populations.
Panel A shows the results for the acp1 target, an additional population of fluorescence measurements is visible with values closely situated to the negatives. Panel B shows the results for the cruA target, an additional population of fluorescence measurements is visible with values closely situated to the positives. Contrary to ‘rain’, the measurements with intermediate fluorescence are not uniformly spread.
Fig 2.
Running a gradient to reduce the effect of co-amplification in digital PCR.
The figure shows the observed droplets for 4 temperatures: 62, 59.8, 58.4, and 56°C (from left to right). All reactions were run at λ ≈ 1. As the annealing temperature is raised the specificity of the reaction increases and efficiency of the co-amplification is reduced, until the undesired droplet population merges with the negative population.
Fig 3.
Illustration of the concept of resolution.
The left-hand figures show the droplet readout, the right-hand figures show the corresponding density plots. tn and tp show the fluorescence positions with the highest density in the negative and positive droplet clouds respectively. wn and wp represent the width of the density peaks at their base.
Fig 4.
Effect of reaction conditions on digital PCR resolution.
Panel A shows the effect of primer concentration. For each of the twelve targets, four primer concentrations were tested (150, 300, 450, and 600 nM). Probe concentrations are such that the primer to probe ratio is the same as for the validated conditions (see S1 Table). Panel B shows the effect of annealing/elongation temperature. For each of the twelve targets, eight temperatures between 62 and 56°C were tested.
Table 1.
Performance parameters for each target under optimised conditions.
Fig 5.
Effect of additional cycles and sonication on the amount of rain.
Round dots represent the results for TC1507. Triangles represent the results for MON88017. Panel A shows the effects of running additional cycles: four conditions of increasing number of cycles were tested (45, 60, 75, and 90 cycles). Panel B shows the effects of sonication: six conditions of increasing sonication were tested (ranging from 0 seconds to 15 seconds in steps of 3 seconds).
Fig 6.
Error as a result of droplet misclassification (positives misclassified as negative).
The effect of misclassification is shown for quantification at three levels (1%, 0.5%, and 0.1%) indicated by different line styles (solid, dashed, and dotted respectively). The black horizontal dashed line represents the amount of misclassification that can be tolerated before 25% error is reached. Panel A shows the results for λ = 3. Panel B shows the results for λ = 1. For the 5% level, 3.3% of the positive droplets can be misclassified before 25% error is reached (1.2% at λ = 1). For the 1% level, this is only 0.73% (0.25%) and for the 0.1% level only 0.075% (0.025%).
Fig 7.
Digital PCR confidence limits.
Panel A shows the relative width of Poisson confidence intervals as a function of the number of target sequences per partition (λ). Panel B and C illustrate how many percent of the λ estimates of a 20 000 partition system contain 25% or more error as calculated via parametric bootstrap for different numbers of repeated analysis (single reaction, duplicate, triplicate, and four repeats). Replicates are averaged to obtain the final estimate. The black lines show the actual percentages obtained, the blue lines show a smoothed version of the curve. The dashed horizontal line represents five percent.
Table 2.
Results from the bootstrap analysis.
Fig 8.
Probability of error in function of sample compartmentalisation.
The variability in positive and negative droplets when only part of the samples is analysed may cause error in the quantification result. The figures show the fraction of results that have more than 25% error (10 000 parametric bootstraps per data point) for different amounts of compartmentalisation and for different percentages of analyte. The yellow area represents 1% of analyte, red 0.5%, and blue 0.1%. The horizontal line represents the 5% criterion. Panel A shows the results for endogene λ = 3 (target λ = 0.03, 0.015, and 0.003) Panel B shows the results for endogene λ = 1 (target λ = 0.01, 0.005, and 0.001).
Fig 9.
Poisson Confidence intervals for 48 independent dilutions of a rare target.
The bottom axis shows the number of copies in the reaction, the top axis shows the number of droplets analysed (as indicated by the green bar-plots). The vertical line indicates the median estimate, confidence intervals that contain this value are coloured black, confidence intervals that do not contain this value are represented as empty boxes. Panels A through D show the decrease in confidence width as a consequence of merging the reactions, the respective CI coverage factors for the panels are: 93.75, 81.25, 75, and 70.83 percent.
Fig 10.
Confidence intervals for 48 independent dilutions based on standard deviation of repeated measurement.
The vertical line indicates the median estimate, confidence intervals that contain this value are coloured black, confidence intervals that do not contain this value are represented as empty boxes. Panel A shows the results for a rare target, Panel B shows the results for an abundant target, and Panel C show the results of their ratio, the respective CI coverage factors for the panels are: 93.75, 93.75, and 97.92 percent.