Target Density downsampling is a method to discard events based on their local density. Please read our introduction to sampling if you have not done so already to learn about your options for calculating local density.

 

Target density downsampling is implemented as originally described in Qui, P et. al. "Extracting a Cellular Hierarchy from High-dimensional Cytometry Data with SPADE", Nature Biotechnology, 29(10):886-891, 2011. The goal is to ensure that the final density of the sampled population falls within a certain density range. Events with local densities below the desired density range are discarded. Events with the local densities within the desired range are kept. Events with local densities above the desired range with are discarded with a probability that depends on how much the local density is greater than the desired range. The higher the local density is compared to the desired range, the more likely it is to be discarded.

 

Target Density Downsampling has several options:

 

Specify Density Percentiles

By default the option is checked, and the user can define how the density-dependent downsampling is performed by entering Min Density Percentile and Target Density Percentile values.

 

If the box is unchecked, Sample Size may be defined instead of these two values.

Sample Size

Indicates how many cells to subsample when performing the transformation. This option becomes available when the Specify Density Percentiles check-box is unchecked.

 

If using this option, the Min Density Percentile and Target Density Percentile will be automatically estimated to select cells.

 

If the specified Sample Size is higher than the number of available events, all available events will be sampled. Note that the number of available events is dictated by the Gate option.

 

The minimum value that can be entered is 10.

Alpha

Alpha is used to calculate the local density. Please refer to the Local Density section of the Sampling Options chapter for more details.

Min Density Percentile

The Min Density Percentile option becomes available when the Specify Density Percentiles check-box is checked.

 

The Min Density Percentile is the percentile within the density distribution of all the cells below which a cell will not be sampled. i.e. if the Min Density Percentile is 1 that means than any cell whose local density is below the 1st percentile of all the local densities will not be sampled.

 

Values range from 0 to 100. The default is 1.

Target Density Percentile

The Target Density Percentile option becomes available when the Specify Density Percentiles check-box is checked.

 

The Target Density Percentile is specified as a percentile within the density distribution of all the cells. If a cell’s local density falls between the density values corresponding to the Min Density Percentile and Target Density Percentile, the cell will be sampled.

 

If the local density of the cell corresponds to a value greater than the one corresponding to the Target Density Percentile, it will be sampled randomly with a probability equal to the (value corresponding to the Target Density Percentile)/Local Density of the cell.

 

Values range from 0 to 100. The default is 20.

Local Density Method

Local Density is further described in the Local Density section of the Sampling Options chapter.

Add Local Density as Output

Check this box to create a new parameter that reports the local density value for each event.

Local Density Output Name

If Add Local Density as Output is checked, the user can specify a name for the new parameter.