About

A Shiny web application for ddPCR analysis. It is part of the twoddpcr Bioconductor package. If you use this package, please cite it.

General Usage

Work your way along the tabs. For more detailed help, see the 'Help' section in each tab's side panel.

Labels

Set the labels for the channels. These will be printed on the plots and the summary tables generated.

Wells in Plate

This shows an overview of the plate. Each of the smaller boxes represents a well in the plate. Droplet amplitudes are plotted for wells that were used.

Droplets in Selected Wells

A density plot of the selected wells.

Classification

The outcome of the chosen classification method.

Plate summary

A summary of the number of droplets in each of the chosen wells. Poisson estimates of the starting numbers of molecules is also provided.

Options

Plot limits

Current training data

If set, the current training data will be shown below. This will be used as training data for the k-nearest neighbour alogrithm.

About

A Shiny web application for ddPCR analysis. It is part of the twoddpcr Bioconductor package. If you use this package, please cite it.

The package and this Shiny app are maintained by Anthony Chiu. It was developed using ddPCR experiments performed by Mahmood Ayub. The project was managed by Caroline Dive, Ged Brady ( Clinical and Experimental Pharmacology Group, Cancer Research UK Manchester Institute ) and Crispin J. Miller ( RNA Biology Group, Cancer Research UK Manchester Institute ).

General Usage

Work your way along the tabs from left-to-right. Help information is available in the side-panel on each tab. Each tab can be summarised as:

  1. Input: Select the dataset.
  2. Select Wells: Choose the wells to use
  3. Classify: Gate the wells.
  4. Results: View the results.
  5. Summary: View/extract the computed figures.

Input

Select the dataset to use. This can be:

  • Loaded from droplet amplitude CSV files. These can be exported from Bio-Rad's QuantaSoft software but must be two channel data; further instructions can be found in the twoddpcr package vignette. Note that empty wells will be ignored.
  • The sample KRAS dataset can be used as a toy example.

The labels for the two targets (in the two channels) can be customised.

  • 'Ch1/Ch2 Label' are used in the plots of the droplet amplitudes.
  • The corresponding 'Abbreviation' field is used in the summary of computed figures. They are prefixed to various column names.

Select Wells

This tab shows an overview of the wells that were loaded. From here, we select the samples/wells to be used:

  • 'All' uses all of the wells.
  • 'Manual Selection' allows the selection of individual wells.

We can choose to view the nonempty wells by changing the 'Plate View' to 'Hide Empty Wells'.

Classify

This tab is used for the gating of the droplets. The 'K-means Clustering' and 'Thresholds' approaches are discussed in the Bioconductor vignette. In general, 'K-means Clustering' should work without any modifications if the data forms four clear clusters; just click 'Run Classification'.

If the clusters are well separated, 'Thresholds' should work too.

K-means Clustering

K-means clustering generally works well but it requires well-selected cluster centres. To do this:

  1. Choose a class from 'Selected Class'.
  2. Click on the density plot to set a new centre for the selected class.

Use the 'Remove This Class' checkbox if there is no cluster.

Note that badly chosen centres could lead to error messages. In this case, try setting centres close to the centres of clusters and remove classes if they are not present.

Thresholds

This works well if the clusters are well separated. This mode sets thresholds to divide the plot into four quadrants. To do this, either

  • Click on the plot to set the new thresholds.
  • Manually change the 'Ch1 Threshold' and 'Ch2 Threshold' figures.
Grid

This mode divides the plot into four regions, each corresponding to one of the four classes. Any droplets not included in these regions are treated as 'rain'. To change the regions:

  1. Choose a class from 'Selected Class'.
  2. Click on the density plot to set a new region for the selected class.
K-Nearest Neighbour

This method requires the use of training data. This method should only be used by experts. The full work-flow is as follows:

  1. To set training data: classify some wells (e.g. one or two wells with little noise) and then 'Set as Training Data' in the 'Results' tab. The resulting training data can be viewed in 'Advanced > View Training'.
  2. The parameter 'k' can be adjusted, i.e. for each droplet, the k-NN algorithm finds the k-nearest training droplets and decides on the classification by majority vote.
  3. 'Use k-NN Confidence' refines the majority required, i.e. if the proportion of the winning class does not exceed the 'Confidence Level', then no classification is assigned.

Results

This tab shows the resulting classification. Which wells to show can be changed under 'Sample to View'.

The ambiguous regions between clusters ('Rain') can be removed using two methods. The 'Mahalanobis' method is recommended with some manual tweaking of the parameters. It fits clusters to ellipses, while the 'Standard Deviation' method is restricted to linear cut-offs.

  • Mahalanobis: This is based on Mahalanobis distance. It fits each cluster to a distribution and removes droplets too far from the cluster centre. This can be adjusted for each cluster with the sliders.
  • Standard Deviation: This works calculated the mean and standard deviation of each cluster in both channels. These two figures are used to find linear cut-offs.
The classification can be exported as a CSV file for later use. It can also be set as training data for the k-nearest neighbour classification method.

Summary

This tab shows a summarised count of the numbers of droplets in each class and the resulting estimates of the concentration of mutants in each sample. It also reports the fractional abundance ('FracAbun') of the channel 1 target. In some use cases, this corresponds to the variant allele frequency.

This summary table can be exported as a CSV file, which can then be imported in a spreadsheet. The results and summary table can also be used to create an HTML report.