GEDI
Gene Expression Dynamics Inspector v2.1
User Manual - Overview
February, 2005
Developed by
at Children’s Hospital Boston,
gedi@childrens.harvard.edu
This software is made available for
non-commercial research use only, with permission from Children's Hospital,
1.
INTRODUCTION
1.1. What is GEDI?
GEDI is a program for the analysis of
high-dimensional data, such as genome-wide profiling of gene expression. It allows
the visualization, inspection and navigation through large sets of such data. It was originally
developed for displaying dynamic data, specifically, multiple
parallel gene expression profile time courses, in order to map the “gene
expression state space”, e.g., following treatment of the same system with an
various drugs. However, it can also effectively help analyze large amounts of static
profiles (e.g. patient samples) and help find patterns of expression without a
priori knowledge of any underlying structure.
1.2. The basic philosophy
The basic idea behind GEDI is fundamentally
different from conventional clustering programs, such as hierarchical
clustering. The goal is not to find “clusters” of co-regulated genes. Data
analysis in GEDI is sample-oriented
rather than gene-oriented, i.e., the object of analysis is an array or sample (e.g., condition, patient, time
point). Although in conventional hierarchical clustering, a “two-way”
clustering is performed, producing both gene clusters and sample clusters, a
given sample as an entity is lost, since it will end up as a branch in a
dendrogram that can become very dense when many samples are involved. In GEDI,
the notion of a sample as an entity is preserved. In contrast, the very idea of
grouping (“clustering”) of samples based on their gene behaviors or of genes
based on their behavior in the various sample is a secondary, “emerging” process.
At the core of GEDI, each sample (= array) is mapped into a “mosaic (“GEDI map”)
which facilitates the recognition of genome-wide patterns, but at the same time
allows the user to zoom-in onto genes of interest given its participation in
the emerging global patterns. Thus, GEDI covers multiple scales of information,
adhering to the spirit of “systems approaches” to present “the whole” as an
integrated entity - yet allowing to directly link system features to “the parts”,
the individual genes.
1.3. How GEDI presents the data
In brief, the output of GEDI is characteristic
mosaic or “GEDI Maps”, a two-dimensional
grid picture for each sample. Each tile in the mosaic represents a “minicluster” of genes (e.g., 10 genes)
whose behavior across the entire set of analyzed samples is highly similar.
Similarly behaving miniclusters in turn are placed in the same neighborhood on
the grid, hence creating a higher-order mosaic pattern. The color of the tile
in a particular mosaic (i.e., SOM grid) represents the expression level of the
genes in that minicluster in that particular sample. The tiles in each mosaic
of the same analysis represent the same genes, thus enabling the direct comparison
of samples (microarrays). In summary, GEDI maps each array into a mosaic
pattern as a well-recognizable, memorizable and characteristic object that
gives each sample an engrammic identity.
1.4. Under the hood of GEDI
The underlying algorithm for creating the mosaics
is an SOM (Self-Organizing Map) algorithm. However, unlike the conventional applications
of SOMs to classify genes or samples into a predetermined number of discrete
groups (“clusters”), GEDI creates a SOMs-based mosaic for every sample. The
learning process is not used to classify genes, but to generate the mosaic
patterns based on miniclusters that allows human eye to find similarity based
on “Gestalt” perception. In other words, GEDI can be thought of as taking the
dots on a microarray that represent expression levels of each gene, grouping together
small groups of genes that behave very similarly over the set of arrays to
miniclusters (dimension reduction) ) and rearranging them so that similar miniclusters
are placed next to each other.
GEDI runs SOMs in two phases. 1st
phase is the rough training phase. 2nd phase is the fine training
phase. For different phases, different parameters can be set in the settings
(see below).
The number of SOM miniclusters clusters
translates into the ‘resolution’ of the mosaic, and we use SOMs with hundreds
of nodes, or ‘mini-clusters’, typically containing 0-20 genes, to create high
resolution mosaic pictures. The color of each tile of the mosaic is determined
by the centroid (approximate to the average of all the genes in that tile) value
of that respective mini-cluster.
Because SOMs place similar genes into the
same neighborhood, coherent and robust pictures emerge that are characteristic
of every sample. Every sample (or array) is associated with one mosaic picture.
Due to the concatenation of the
samples and time courses in the input data matrix (see data file preparation
documents), every tile of the mosaic in each of the mosaics corresponds to the
same group of genes across the sample
GEDI can analyze two different types of gene expression profile data:
|
n
Dynamic analysis: Analysis of multiple parallel time-series
data, allowing the comparison of multiple high-dimensional time courses, e.g.
following treatments with various drugs. A mosaic then represents a state in
the gene expression state space, and its trajectory through state space is
reflected in the change of the mosaic patterns which can be displayed as a
movie. n
Static analysis: Analysis of non-time-series data, e.g. expression profiles of tumor
samples from different patients, or comparison between normal or disease
sample, etc. Because of its original design for parallel time series data and
the concept of sample-oriented analysis, GEDI allows efficient navigation
through “sample space”, facilitating quick visual comparison of individual
samples in a large set of profiles, similar to sorting through a large stack
of trading baseball cards. |
à The
input data format, which can be obtained by simple manual modification of the
expression data spreadsheets (see 3.1.) determines whether GEDI will analyze in
the Dynamic or Static data mode.
1.5.Screen Shot
GEDI displays the analysis name and the
data set name on the title bar of the program. The main area of GEDI is the
working areas; it is organized into 4 Tabs:
·
Welcome: This is the welcome screen when you just
open GEDI.
·
Data: This is to display the information about
the dataset.
·
GEDI-Map/Movie: This is the main working area to view
GEDI Maps and detailed information about genes, etc.
·
Display/Calculation: This allows user to selectively display a
subset of the samples, or to calculate “Average Maps” or “Difference Maps” from
selected GEDI mosaics.
Above the working area are the menu items
and some tool bar buttons for commonly used functions.
1.6.Basic Capabilities and Feature
n
Sample clustering: Every sample is
translated into a distinct mosaic, yielding a stack of mosaics. The user can
browse though the stacks and can choose to display selected samples The mosaics
can be saved independently as jpeg files, and used for presentation and printing
using standard graphics programs, such as PowerPoint or Corel Draw.
n
Gene clustering: Genes that behave similarly across the set
of samples are located in tiles in the same region of the grid. Tightly
co-regulated genes are assigned to the same tiles. Thus, genes that behave
similarly in a subset of the samples, and distinctively between samples, will
spring to eye, e.g., “as a red island in a sea of blue”. Clicking on the spots
will show the genes in the gene list window.
n
Sample comparison: Static samples can be selected and used to
calculate either “Average maps”,
representing the “average mosaic” that displays the average value for every
tile from the various samples, e.g., of a group of selected samples
representing same diagnosis; or “Difference
maps” by subtracting two mosaics tile-wise to facilitate sample comparison.
n
Animation: If a stack of mosaic contains a group of
sample representing time points of a time course, the stack can be animated to
show the change of the transcriptomes, allowing recognition of coherent dynamic
patterns across the profile.
n
Analysis of individual genes: User can retrieve the genes associated
with features in the mosaic patterns. Gene
description of a gene selected based on a feature in a GEDI map for one
sample and along with its behavior in
other samples can be displayed by clicking on the tiles, which displays the
genes contained in that respective minicluster, and by clicking on the specific
gene in the list. Alternatively, a
specific gene name can be search for by entering its name in the search
window, which will display its location on the mosaics.
1.7.The work flow and work environment of GEDI
The GEDI environment is simple. It consists
of one single user interface window containing various, fixed displays. Working
with GEDI consists of the following steps:
n
Loading the data as one text file. The data will need to be
in a simple predefined format, see section 3.1., based on which GEDI will
recognize whether the data represent static or dynamic data.
n
Setting the parameters. This step is optional, the default
settings work well. Settings pertain to both SOM parameters (mosaic resolution,
learning) as well as the display (coloring, grid shape, etc)
n
Create mosaic stacks (static analysis) or
movie (dynamic analysis).
n
Visual inspection of the results in the dynamic user interface: Watching
movies of the animated mosaics, browsing through stacks of mosaics, selecting
interesting samples for direct comparison, retrieving gene names based on
interesting patterns or locating genes by name in the patterns
n
Export results as Html pages, Jpeg images, or gene lists.
The session with results can also be saved.
2. REQUIREMENTS AND INSTALLATION
2.1. Minimum
Requirements
The minimum hardware requirements needed to run GEDI:
Video display capable of displaying at least
1024 by 768 Pixels
The software
requirements needed to run GEDI:
GEDI version 2 is developed using Java technology. It is portable on all the platforms that support Java. Check with your computer administrator before you install Java. You can download it from Sun Microsystems, Inc. (http://java.sun.com/j2se/1.4.2/download.html).
2.2. Installing GEDI
2.2.1 Download
and unzip the GEDIv21.zip file.
Download the GEDI file from GEDI Website: http://www.childrenshospital.org/research/ingber/GEDI/download.htm
Inside this zip file, you will find the
GEDI folder. Place the folder and all of its contents to "C:\Program
Files" (Unix user can choose a location in your user directory). We will
refer to this directory as "GEDI Root Directory".
2.2.2 Initial
setup
Ü First, start GEDI by double-clicking on the
program icon “gedi21.jar”. After you start GEDI, go to "Settings" menu to set the default
files/directory settings, point them to appropriate directories in the GEDI
Root Directory.
3. USING GEDI
3.1. Preparing
the Input Data
The input data format is as described in GEDI Input Data Format.
3.2. Running
GEDI
3.2.1. Starting
GEDI
After starting up, GEDI will display the opening interface in the “Welcome” tab.
Ü To begin a new analysis, click on “Start new GEDI session” button (or use
the same common in the “File” pull down menu. A dialogue window will open allowing you to
load your data file (continue on section 3.2.2).
GEDI allows you to save an entire analysis session. You can click on “Load Saved GEDI session” button to
browse and load your previously saved GEDI session. Or you can simply open the
GEDI session that you saved or open last time. The file path to latest visited
session file is also displayed.
3.2.2. Load your
Data
Ü Locate and
select your file which has been manually preformatted (see 3.1.) in the “Load data file” window, and click “Open”.
If the data file format is correct, after
reading GEDI will go to the “Data”
screen, display a summary of your data, and give you options to either run the analysis or change settings.
3.2.3. Enter
parameter values
At this point, GEDI will allow you to
change parameters pertaining to the SOM
algorithm as well as other options, such as display.
Ü This step can be skipped by clicking “Analyze with current settings” to
proceed with the analysis of the data.
Ü To change the parameters, click “Change settings”. You can also access
the settings popup window from the “Settings” toolbar button/menu item. This
will open the following window:
This GEDI Settings window has 5 tabs:
n
SOM: Basic parameters for SOM analysis. For expert users,
click on the “Advanced settings” will
go to the “Advanced” tab for tuning basic SOM parameters, such as grid size
n
Map/Movie Properties: Define how the GEDI maps will look like.
n
File location: Sets working file directories for GEDI
n
Import/Export: Import or Export GEDI Settings as XML
file.
n
Advanced: Advanced setting for SOM parameters.
Note that for settings in the SOM and Advanced tab, after changing the parameters, you will need to rerun
the analysis for them to take effect. For the other tabs, the change will take
effect immediately after confirming the change.
If the you check “Save to system” checkbox and click “OK” or “Apply” to confirm
the change, the settings will be saved to the system and be loaded next time
you open GEDI.
A detailed description of each of the
parameters is provided in a table in
GEDI_Settings.
3.2.4. Generate
the Mosaics
Ü Once the desired parameters have been set,
simply click the “SOM Analysis”
button/menu to start analysis
Depending on the grid size, training
iterations, and the computer speed, GEDI will take between minutes to hours to calculate
and generate the mosaics. The longest step is typically “Generating the SOM” as will be indicated in the status bar window,
so have patience if this procedure seems to last abnormally long. The status
bar displays the percentage of analysis that is complete. Sometimes, it is
advisable to estimate the calculation time by running the analysis with small numbers
for grid size and training iterations first.
When the SOM Analysis is complete, GEDI
will automatically display the result GEDI Map in the GEDI-Map/Movie Tab.
3.3. Navigating
through GEDI
Once you have created visualization, you
can explore the features and genes of interest underlying the
visualization. This is be performed
within GEDI–Map/Movie Tab.
Following four panels display the results: On
the upper panel, there are two maps: GEDI Maps (see 3.3.1.) and Gene
Density Map (3.3.2.). On the lower panel, there are the “Selected
Gene List” (3.3.3.) and “Gene Expression Plot” (3.3.4.). Note: On monitors not set to the
resolution 1024 x 768 pixels, you may have to resize or maximize the GEDI
window for all panels to appear.
3.3.1 GEDI Maps
This area displays the GEDI mosaics from
the SOMs analysis. The number of maps is corresponds to the number of samples
(columns) in the input data. You can choose to display only a subset of samples
by making the selection in “Display/Calculation”
Tab (3.4.3.).
Each tile on the map was assigned a color
based on its centroid value of the
expression level of the represented minicluster found in the SOM. This value
roughly represents the average expression of all the genes in that minicluster
in that respective sample. The mapping between a color and its corresponding
value is displayed in the color bar. Basically “Red” represents the highest expression
level “Blue” the lowest expression level.
For interactive browsing through the
miniclusters in the GEDI Map mosaics, note the difference between two statuses
that each tile can have: selected and active.
Ü To select the tile on the mosaic, you
can click on a tile. Click again will
de-select the tile. You can also press
the mouse button to draw a rectangle to select/de-select an area. The selected area will have a border
around it. A selected tile will have its tab in the Selected Gene List, under which the genes in the selected tiles are
listed.
The diamond shape marker indicates the active tile
which is usually the last selected (clicked) file, unless you activate a tile
without selecting it (see below). The active tile is the tile whose genes are
displayed as the open list in the “Selected
Gene List” window. The gene annotations (specified in the data file) for
the active tile will be displayed in “Selected
Gene List” box. The Gene Expression Plot
for this tile will be display in the lower right panel. The gene lists of
selected but not active tiles appear as inactive tabs in the Selected Gene List window.
Note that the tile status is independent of
whether the tile is selected or not. This allows you to inspect one of several
selected tiles and also to display the gene list of a non-selected tile.
Ü To activate a tile without selecting it, i.e., to display its gene list content, move
the diamond marker using the arrow keys on the key board.
Ü When a tile is active, hitting “space” key in
the keyboard will toggle the tile to be selected/de-selected.
For dynamic analysis, mosaics from the time
course sample are grouped together to a stack and displayed as a window. Below
each stack of mosaics is a slider. Each
slider will allow you to browse through the mosaics in the above window.
Ü If there is more than one window of mosaics
(when comparing multiple time courses), their sliders can be synchronized by
checking the Sync (Synchronize Scroll)
check boxes adjacent to each slider.
This feature enables scrolling through many stacks simultaneously.
Ü Press the “Watch Movie” button to sequentially display each mosaic in its
window as a movie.
(For static analysis, each window only has
one mosaic (i.e., is not a stack). Therefore, there is no slider and Sync check box. )
GEDI assigns a representative color for
each mosaic window in the form of a color header label located above the mosaic
window. This header serves to link each mosaic window with all other graphics,
such as the Gene Expression plot.
3.3.2 Gene
density map
The Gene
Density Map displays the number of genes assigned to every particular tile
(minicluster). The value of the gene number is mapped to a color. The gene
density map helps optimize the grid size for the SOM. Ideally, the gene density
should be evenly distributed, with 1-20 genes per minicluster and there should be
minimal or no empty tiles. Occasionally (depending on data), some tiles will
have exceptionally high number of genes (100) – this is often difficult to
avoid.
The gene density map is “active”, i.e.,
genes also can be selected in this map by clicking the tiles. Mouse click and keyboard input have the same
behavior as in the GEDI maps.
3.3.3 Selected
gene list
Genes selected in the GEDI maps will be
listed in the Selection Gene List window.
All the genes on each selected tile will be organized in separate tabs within
this window. Each of the individual tabs represents a tile (minicluster) and is
labeled with the mosaic grid coordinates of the tile and the total gene number
in this tile.
Ü Click
on an individual gene in the gene list to mark the tile where it belongs as
active tile in the GEDI Maps and to highlight the gene in
the Expression plot window on the
right.
Ü To search
for a gene using its name and find any occurrence of the search target found in
the gene description entered in the input data file, enter the search string in
the Search window, and click the Search Gene button (magnifying glass).
The
matching criterion is “exact match”
with any part of the gene description. The result will be listed in a tab
labeled as “Search”. Select the gene again
will highlight the tile that contains the gene in the GEDI Map and in the Gene Expression
plot.
Note: the search tool is not case sensitive.
Other buttons in the “Selected Gene List”
panel:
Ü “Filter in” (“and Filter out”). GEDI
allows a reanalysis of the data with just a subset of genes that are manually
selected or excluded in this window. To
keep only the selected genes in the gene
lists of the selected tiles, click the “Filter
in” button, or to exclude the selected genes, click “Filter out”. This will create a subset of the dataset in the memory
(but will not write in the data file), that can be subjected to a new GEDI
analysis. The current analysis will be lost; therefore, a confirmation dialogue box will appear.
Ü “Clear markers” button: de-select all
the tiles and clear the gene list.
Ü “Save marked nodes” button: save the
content of the selected tiles as a text file.
3.3.4 Gene Expression
Plot
This
window provides graphical information on individual genes through sample space,
i.e., the information used for the SOM to define the miniclusters. Thus, this
information can be used to gauge the quality of the SOM.
For dynamic analysis, the gene expression
time courses will be plotted as curves (horizontal axis = time; vertical axis =
scaled gene expression value). The thick solid lines on the plotted
graph represent the gene highlighted in the Selected
Gene List. The thin solid lines
correspond to the genes from the same tile, but not highlighted in the list. The
thick
dashed lines correspond to the centroid
value associated with the selected tile.
Ü Using the checkboxes on the right you can
choose which of the curves to display.
For static analysis, gene expression
will be plotted as bar chart. Each bar represents a sample, with the height of the thick bar representing the centroid
value of the active tile for that
sample. Selecting a bar will highlight the GEDI Map representing the same sample.
If a gene in the gene list is
highlighted, it will be plotted as thinner solid color bar superimposed
on the centroid bars for all the samples.
3.4 Calculating
Mean and Difference
Sometimes in static analysis the number of
samples is very large and contains groups of nominally identical samples
(repeats). To facilitate navigation through the many samples, GEDI can
selectively display GEDI maps. Even better, GEDI allows you to compact the
sample space by generating an “average”
GEDI map in which the color now represents the values of the centroid, averaged
over the corresponding centroids (tiles) of a group of selected samples. You
can also calculate the difference between pairs of GEDI maps and present the
result in a new map. These functions can be performed in the “Display/Calculation” Tab.
In
the “Display/Calculation” Tab, all the
original samples will be listed with a checkbox on the left hand side, in the “Original Samples” panel. The calculated
result will be listed on the right panel. The calculated result can be selected
like the original samples for display or for further calculation.
3.4.1 GEDI
Average Map
Ü To calculate the average map of a group of samples, select
the samples of interest by checking the
boxes next to each sample. Then select the radio button below “Calculating
average…” and give it a name. After clicking on the “Calculate” button, the result will be listed as a new sample in the
“Calculated Results” Panel.
3.4.2 GEDI Difference
Map
You
can calculate either the difference between any two samples or multiple
differences of pairs of samples where each sample is subtracted by the same
reference sample.
Ü To calculate the difference between two samples, you
need to first select the radio button
“Calculate difference ... ”. Then
first select the “test sample” (or the
samples) by checking the checkbox in the sample lists. Now select the reference sample from the drop down combo box. This reference
sample will be subtracted from all the checked “test samples”. If multiple
checkboxes are selected, the results will display the difference maps for each
of these subtracted test samples. Only one reference sample can be selected.
3.4.3 Selective Display
Ü You can select to display any combination of
original samples and calculated results. After making the selection in the
Check boxes, click on the “Display button”
to return to the GEDI – Map/Movie Tab
which will display the selected GEDI Maps.
3.5 Exporting Results
Ü You can export the analysis results in many
formats. They can be accessed from the menu “File”à“Export results”.
3.5.1 Export
GEDI Maps
The GEDI Maps that are selected (3.4.3.) to be displayed in the “GEDI Map/Movie” Tab can be exported as JPEG files. You will be asked to select a name and location for the directory.
An HTML page will be generated to organize the JPEG files and allow quick browsing through all the GEDI Maps; some important GEDI setting parameters will also be listed on the HTML page. All the parameters will be saved as a XML file, allowing user to import back to GEDI for the same settings. All file will be saved inside the selected directory.
For static
analysis, each sample will be saved as a JPEG. For dynamic analysis each time point (movie frame) in each selected
sample will be saved as a JPEG.
3.5.2 Export Map
Centroids
This will export the centroid value for all the GEDI Maps. The format is shown in
following table. Each column is a sample map. Each row is a tile on the map.
|
}$ |
S1 |
S2 |
S3 |
S4 |
|
1,1 |
-0.85 |
-0.51 |
-0.57 |
-0.93 |
|
1,2 |
-0.96 |
-0.62 |
-0.73 |
-0.91 |
|
1,3 |
-1.03 |
-0.74 |
-0.86 |
-0.93 |
Another format of map centroids of map
centroids is also saved, with the data being transposed. Therefore each column
is a tile; each row is a sample map.
|
}$ |
1,1 |
1,2 |
1,3 |
1,4 |
|
S1 |
-0.85 |
-0.96 |
-1.03 |
-1.04 |
|
S2 |
-0.51 |
-0.62 |
-0.74 |
-0.92 |
|
S3 |
-0.57 |
-0.73 |
-0.86 |
-1.06 |
3.5.3 Export
Gene Assignment List
This will export the genes on each tile as a list. The genes on the same tile will be
group together. First column is the gene name, second and third column are the
coordinates of the tile that contains this gene.
|
Genes
- from AML_ALL |
XCoord |
YCoord |
|
PRKCQ Protein kinase … |
1 |
1 |
|
TCRB
T-cell receptor, … |
1 |
1 |
|
Sterol
regulatory element … |
1 |
1 |
|
CD3G
antigen,… |
1 |
1 |
3.5.4 Export
Density Map value
This function will export the number of genes on each tile as a
matrix.
3.5.5 Export
Marked Tiles
This function is similar to the gene
assignment list, but only with the selected
tiles.
©2005 Children’s Hospital, Boston.