User Manual

Installation

Download ACCENSE.

Mac OS X

After downloading the installer, double click the .dmg file and drag ACCENSE.app from the dmg file into your Application folder

Windows

After downloading the installer, double click the .msi file and follow the instructions on the screen.

Before Start

ACCENSE requires .csv and .fcs as input. Multiple files input is also available if all the files contain identical channels. Once input files are selected, it allows customization on choosing cell numbers and channels.

Dimensionality Reduction

ACCENSE employs t-Distributed Stochastic Neighbor Embedding (t-SNE) for dimensionality reduction [tSNE2008]. Two versions of t-SNE are provided, in which Standard-tSNE is designed for the small data sets; whearas Barnes-Hut-SNE is designed to run on large data sets (data number > 5000).

The steps below introduce how to run the function in ACCENSE.

  1. Open ACCENSE. Click Open button to open Input Window.
_images/mainwin.png
  1. Click Browse to choose the input files. The cell numbers and channels are shown once the files have been selected. Choose cell number for analysis from each file, or use the down sampling to select a total number for all files. Click to choose the channels for analysis. Click Import after the configuration.
_images/filterwin.png
  1. On Main Window, choose dimensionality reduction function between Standard-tSNE and Barnes-Hut-SNE.
  2. Choose Perplexity that configures how well results that the dimensionality reduction wants to obtain. Perplexity is set to 30 by default, which is a experienced value suitable for most of the case.
  3. Select the directory for saving Output.
  4. Click Run to start the computation. Logs on of the running status are shown. Next button will be enabled when it is accomplished, which navigates to Visualization.
_images/statuswin.png

Note

The error during the log is a parameter of tSNE calculation.

Visualization

The result of Dimensionality Reduction, a set of 2D plots, will be shown in the visualization window. The figure below shows a screen shot of it.

_images/visualwin.png

There are two ways opening the visualization window:

  1. Run Dimensionality Reduction. Click Next once the computation accomplished.
  2. Or, choose Cell classification and visualization on the main window. Then click Open button to select the ACCENSE Output file.
_images/reloadwin.png

Basic operations

Zoom, Pan and Home buttons are provided for operations of zoom in, pan and back to the initial, respectively. Save button saves the image to a PNG file. Grid check box provides an option to switch on/off the grid.

_images/figframe.png

Figure Control Panel

Coloring

  1. Basic Coloring: Plots are simply colored by selection.
  2. Channel coloring: Coloring the plots based on the expression of chosen channel with selected colour.
  3. Cell classification and coloring: Coloring the plots based on the classified populations.

File fiter

Click Select button besides Filter label to show the filter window. Then, tick the file name to switch on/off plots of the file.

Cell Classification

ACCENSE provides two clustering methods, K-means and DBSCAN, to automate the classification of cellular sub-populations. To run the compuation, choose the method and click Calculate button. Click the button again to re-run it if you change the parameters.

The obtained sub-populations will be shown in different colour. Each sub-population is also highlighted by index.

K-means

This is an implementation of the algorithm proposed in [K-means2003]. It groups cells into sub-populations with similar protein expression by specifying a standard statistic significance level, instead of a predefined expected populations number.

The populations number can be controlled by changing the significance level parameter. The smaller the parameter is set, the less populations is obtained.

DBSCAN

Density-based spatial clustering of applications with noise (DBSCAN) can find arbitrarily shaped clusters. In the case that the population is not under norm distribution, it obtains better classification.

DBSCAN requires two parameters: eps and min points. For setting the parameters we suggest to read the explanation at http://en.wikipedia.org/wiki/DBSCAN#Preliminaries.

Output

When it is accomplished, an output file accense_output.csv, is saved automatically to the output directory. Format of the output file is shown in the table below. The first column is the input file name. The last three columns are results of Dimensionality Reduction and Cell Classification. The other columns are the channels expressions of input. The chosen channels are marked by *.

file channel_1 *channel_2 ... channel_n ... SNEx SNEy population

The population column contains the index of sub-populations. The smallest index is 1 if it is computed by K-means. If it is computed by DBSCAN, it may contain an index 0, which indicates that the belonging population is noise.

It can also exports each population to a separate FCS file by tick box Export sub-populations as FCS-files. The files are saved in “populations” folder under the output directory. The file name is “population-N.fcs” where N is the population ID. Each fcs file contains the channels expressions of input and results of Dimensionality Reduction.

Reference

[K-means2003]
  1. Hamerly and C. Elkan. Learning the K in K-means. Proc. NIPS, pp.281 -288 2003
[tSNE2008]L.J.P. van der Maaten and G.E. Hinton. Visualizing High-Dimensional Data Using t-SNE. Journal of Machine Learning Research 9(Nov):2579-2605, 2008.