What is new in Release 5.3 of ArrayMiner® (12/Feb/2006)

Clustering Algorithm
- Fast stop
- Windows version ported to 32 bits
- Peculiar data problem fix

Graphical Interface
- Mac OSX Tiger with G5 problems fixed
- GeneSpring data column reordering problem fixed
- Windows JPEG save problem fixed

User Interaction
- Minimize all application available on Windows
- Data file access easier on Windows

Clustering Algorithm

Fast stop

The clustering algorithm now offers a "fast stop": whatever the size of your data, the algorithm stops immediately upon request, instead of waiting for the end of the current iteration. When restarted, the algorithm resumes as if no interruption uccured

Windows version ported to 32 bits

On Windows, the software underwent a full port to 32 bits, resulting in a significant speedup of the clustering algorithm on 32 bits operating systems, such as Windows XP

Peculiar data problem fix

On some quite peculiar data, the clustering algorithm could end up with a completely flat average profile for a cluster, which caused problems under Pearson or Correlation. This was fixed

Graphical Interface

Mac OSX Tiger with G5 problems fixed

On Mac OSX, some low-level GUI problems when running Tiger on a G5 machine were fixed

GeneSpring data column reordering problem fixed

A problem in the mechanism of reordering of columns in GeneSpring data was fixed

Windows JPEG save problem fixed

On Windows, a warning is now issued on failure to save a too large JPEG image

User Interaction

Minimize all application available on Windows

On Windows, pushing F11 now minimizes the whole application

Data file access easier on Windows

On Windows, right-clicking the status bar in the bottom of the Toolbar Button Window brings up a popup menu relative to the data file.

What is new in Release 5.2 of ArrayMiner® (23/Jan/2004)

Graphical Interface
- Project notes available

User Interaction
- Richer web publishing
- Easier license installation

Graphical Interface

You can now add notes to your project

You can now in a single click generate html pages with 3D interactive figures thanks to our Java3D applet technology and use the resulting pages in your website to disseminate the results of your work.(see a running example of the technology here)

User Interaction

Richer web publishing

Saving the clustering solution as html lets you now link the resulting website to an interactive 3D view of your data. New options let you choose to add profile pictures to the html pages.

Easier license installation

A new interface has been added to let you directly paste your license instead of loading it from file.

What is new in Release 5.1 of ArrayMiner® (16/Oct/2003)

Graphical Interface
- Publishing interactive 3D figures on the web added
- Publish a whole classification as a website

User Interaction
- A textual description of the clustering process can now be easily pasted into your publications
- Better support for foreign languages in data input


Graphical Interface

Publishing interactive 3D figures on the web added

You can now in a single click generate html pages with 3D interactive figures thanks to our Java3D applet technology and use the resulting pages in your website to disseminate the results of your work.(see a running example of the technology here)

Publish a whole classification as a website

A new save format has been added for the clustering module, allowing you to save the whole classification as a complete linked website.

User Interaction

A textual description of the clustering process can now be easily pasted into your publications

A button has been added to the clustering info window allowing you to copy into the clipboard information about the clustering parameters, which you can then easily paste into your publications

Better support for foreign languages in data input

The data importer and the copy data from clipboard option now better handle various number formats, including foreign languages such as Finnish and French.

What is new in Release 5.0 of ArrayMiner® (16/Sep/2003)

User Interaction
- Project Explorer window added

Clustering Algorithm
- Gaussian Clustering algorithm improved

User Interaction

Project Explorer window added

This window lets you filter, merge, exclude experiments easily and offers a more intuitive way of handling your data.

Clustering Algorithm

Gaussian Clustering algorithm improved

The precision of the complex computations involved in the algorithm has been increased, allowing data with very high numbers of experiments to be clustered safely.

What is new in Release 4.1 of ArrayMiner® (1/Jun/2003)

Graphical Interface
- Experiment Tree Window
- Fish Eye View
- Raw data window improved

User Interaction
- Profile inspector window
- Improved classmarking filtering data introduction
- Possibility to create a clustering tree
- Easier data importing
- Unlimited window size
- Improvement of the graphical interface performance, usability and stability on PCs

Graphical Interface

Experiment Tree Window

In addition to ArrayMiner unique data mining algorithm, an experiment tree window has been added. This gives you an intuitive way to understand how your experiments match each other.

Fish Eye View

The fish eye view is another way of displaying experiment trees letting you more easily understand the tree structure.

Raw data window improved

The raw data window has been redesigned. You can now check your data interpretation with a table of values or with a graphical representation of each gene before and after normalization.

User Interaction

Profile inspector window

The profile inspector window lets you display all profiles in a cluster before or after normalization. It also lets you easily compare two clusters. It is in particular accessible in the classification compare window.

Improved classmarking filtering data introduction

A new filtering window lets you change all the filtering values of the classmarker module simultaneously, letting you edit the full parameters without waiting for the normalization to end.

Possibility to create a clustering tree

In the clustering module you can create a clustering tree with the average profiles of all clusters.

Easier data importing

The ArrayMiner converter application has been removed and replaced by an import data command in the file menu. The new importer is far faster than the converter application and can handle more formats.

Unlimited window size

Most windows are no longer limited in size.

Improvement of the graphical interface performance, usability and stability on PCs

The overall graphical performance of the PCs has been improved. Actions like scrolling or resizing windows are much faster than in previous versions. Keeping mouse down on the scrollbar buttons is now correctly handled. Some rare issues which yielded white windows on some machines had been corrected.

What is new in Release 4.0 of ArrayMiner® (22/Apr/2003)

Graphical Interface
- Improved graphics publishing
- Customizable colors and heat map gradients
- PCA vector profile visualization

User Interaction
- New formats to save data and results
- Classification Compare Window improved
- Coloring schemes improved
- Easier introduction of number of clusters
- Improved cluster export to clipboard

Graphical Interface

Improved graphics publishing

All the ArrayMiner views are now publishable as bitmaps in various formats and resolutions. A heat map publisher in the clustering module lets you easily configure the heat map output.

Customizable colors and heat map gradients

The cluster colors and heat map gradient colors are now editable via a preference panel.

PCA vector profile visualization

PCA vector profiles can be viewed and published in the clustering module. A PCA window has been added to browse the PCA axes of your data in the clustering module.

User Interaction

New formats to save data and results

You can now customize the way you want to save your data. It is also possible to specify more precisely which information you want to save.

Classification Compare Window improved

The class compare window is now much more customizable. You can edit classification and cluster names within the window. It is also now possible to display in the interface any classification loaded in the Classification Compare window. You can easily introduce a new classification from a simple gene list in the clipboard.

Coloring schemes improved

It is now possible to introduce a new coloring scheme from a simple gene list in the clipboard.

Easier number of clusters introduction

The number of clusters sought by the clustering process can be changed at any time which lets you easily compare results with different number of clusters with the Classification Compare window.

Improved cluster export to clipboard

You can now easily customize which cluster information should be copied to the clipboard via an exhaustive preference panel.

What is new in Release 3.3 of ArrayMiner® (18/Feb/2003)

Graphical Interface
- Internet queries interface
- Cluster names available
- Improved graphic export

User Interaction
- GeneSpring users
    * Classification import from GeneSpring
    * Gene descriptions and Internet queries inherited

Graphical Interface

Internet queries interface

Users may now specify internet queries and search the Internet inside ArrayMiner.

Cluster names available

It is now possible to name the clusters. The names appear in the graphical windows and are sent back to GeneSpring.

Improved graphic export

You can now specify the size and file settings when exporting the contents of various areas in the graphical interface. Popular image formats are supported for easier publishing.

User Interaction

GeneSpring users

You can now directly import GeneSpring classifications into the Classification Compare Window or set them as coloring schemes. (GeneSpring 5.0 required). You can now use the new "Refresh Annotations" option in the "External Programs/ArrayMiner" menu of GeneSpring to send the description fields and your Internet queries to ArrayMiner.

What is new in Release 3.2 of ArrayMiner® (26/Nov/2002)

Graphical Interface
- Graphical interface has been fully redesigned


Graphical Interface

New density graph mode for the 3D and 2D views. Redesigned window header in the clustering mode to facilitate inter cluster navigation. Version check option to insure that you always get the latest version. Faster search window.

What is new in Release 3.0 of ArrayMiner®

New functionality
- New class marking and prediction functionality available.

Clustering Algorithm
- Dramatic speedup on large datasets.

User Interaction
- Data column reordering under GeneSpring accepts column aggregation.


New functionality

New class marking and prediction functionality available

The optional functionality ClassMarker is now available in ArrayMiner. This class marking and prediction tool allows researchers to identify genes discriminating between samples in different classes, such as various cancers, different tissues, etc., and to classify unknown samples on the basis of those marker genes. A step-by-step tutorial and detailed online help facilitate the learning of the tool.

Clustering Algorithm

Dramatic speedup on large datasets

By means of advanced programming techniques, the clustering algorithm now runs significantly faster. Supplying the same high-quality clustering results as before, the algorithm is ten times or more faster on large datasets.

User Interaction

Data column reordering under GeneSpring accepts column aggregation

In addition to the reordering of data columns to match those in GeneSpring®, ArrayMiner is now also able to aggregate columns when necessary.

What is new in Release 2.6 of ArrayMiner™

Installation
- Better support for distributed installations of GeneSpring®.

User Interaction
- New seamless GeneSpring interface on the Macintosh.
- Data column reordering under GeneSpring.
- New ArrayMiner launcher for GeneSpring

Installation

Better support for distributed installations of GeneSpring®

ArrayMiner now correctly installs in GeneSpring environments where data have been moved from the default GeneSpring location elsewhere, e.g. to a remote server. This situation was previously handled by a manual copy of the data/Programs folder to the remote location.

User Interaction

New seamless GeneSpring interface on the Macintosh

ArrayMiner takes advantage of GeneSpring 5.0’s availability under Macintosh OSX, offering the same seamless data exchange as on Windows, i.e. with no intermediate file manipulation involved.

Data column reordering under GeneSpring

A new functionality works around a glitch in GeneSpring’s External Program Interface that sometimes causes ArrayMiner to display data columns (attributes) in an order different from GeneSpring’s. The proper column order is now easily copied from GeneSpring. In addition, the manipulation needs to be done only once for any GeneSpring Experiment, since the column order is saved and automatically retrieved when necessary. This was previously handled by data exchange via the clipboard.

New ArrayMiner launcher for GeneSpring

The ArrayMiner launcher (Java wrappers) has been completely redesigned, offering additional setup capabilities for atypical GeneSpring installations.

Clustering Algorithm
- Better support of high numbers of columns in the Gaussian model.

Graphical Interface
- New welcome dialog.

User Interaction
- Data import from clipboard.
- Interactive help system.

Clustering Algorithm

The gaussian model of clustering can now handle files with more than 60 columns. In the previous version, this kind of files could yield to a single cluster due to precision issues.

Graphical Interface

A welcome message lets you choose different start points and lets you automatically load the demo file.

User Interaction

The "Get data from clipboard" option lets you select and copy your data in a software like Excel and retrieve it directly in ArrayMiner.

A new interactive help system using hypertext help file is accessible via the "Help Menu", and from important points in the software.

Clustering Algorithm
- More explicit selection of clustering model
- 2D and 3D PCA view
- Improved Macintosh interface

Clustering Algorithm
- Improved interface for PC GeneSpring users

Graphical Interface

More explicit selection of clustering model

The selection of clustering model between the classic (minimal variance) criterion and the new Gaussian Mixture model has been made more explicit with a new window specific to that choice.

2D and 3D PCA view

The popular 3D view now offers a PCA (Principal Component Analysis) mode that displays the data in eigenvector axes instead of the usual experiment axes. The same is also available in the 2D view.

Improved Macintosh interface

The interface of the Macintosh port of ArrayMiner has been improved to better follow the Macintosh interface guidelines.

User Interaction

Improved interface for PC GeneSpring users

PC users who run ArrayMiner from GeneSpring™ by Silicon Genetics, can now restart the clustering process with a new set of parameters without leaving ArrayMiner, sparing the process of data exchange between GeneSpring and ArrayMiner.

2.3

Clustering Algorithm
- Gaussian algorithm improved

Clustering Algorithm

Gaussian algorithm improved

The Gaussian Mixtures clustering algorithm has been improved, yielding a faster convergence to high-quality solutions on some data.

Clustering Algorithm
- Gaussian mixtures based on GGA instead of EM

Clustering Algorithm

Gaussian mixtures based on GGA instead of EM

The Gaussian Mixtures clustering, previously based on the EM algorithm, has been completely reworked and is now based on Optimal Design's GGA. This results in a dramatic increase of efficiency of the algorithm, due to the power of the GGA on the one hand, and a substantial body of programming "tricks" made possible by the use of the GGA on the other hand.

What is new in Release 2.1 of ArrayMiner

Clustering Algorithm
- Outliers (noise) detection

Clustering Algorithm

Outliers (noise) detection

The Gaussian Mixtures algorithm was enhanced to detect outliers (noise). Noise is modeled as an additional uniform probability distribution that "competes" with the Gaussians. This adds stability to the algorithm, because it allows it to "abandon" data points that are impossible to cluster with the other profiles, instead of trying to classify them at any cost. On the other hand, it allows for detection of unique expression profile patterns in the data, that may constitute the interesting discovery in the data set.

What is new in Release 2.0 of ArrayMiner

Graphical Interface
- Translucent 3D solids supported on graphic cards not supporting 15-bits color
Clustering Algorithm
- Gaussian Mixtures clustering

Graphical Interface

Translucent 3D solids supported on graphic cards not supporting 15-bits color

The translucent 3D solids were displayed incorrectly on some graphic cards not supporting the 15-bit color depth. This has been corrected.

Clustering Algorithm

Gaussian Mixtures clustering

A new breed of clustering algorithm is being introduced, based on the Expectation Maximization algorithm. In this approach, instead of clustering for minimal total variance of the clusters, a set of Gaussian distributions is fitted to the data. This allows in particular to detect adjacent clusters of unequal dispersion (variance), which is impossible to achieve with the more standard distance-based clustering.

Translucent 3D solids representing clusters

The popular 3D solids (convex hulls) interface is further enhanced by showing the solids translucent (see-through). This option adds even more "feeling" of the spatial relationships among clusters.

Coloring by external classifications

A highly efficient way of comparing classifications consists of coloring the spheres representing the gene profiles in the 3D cube with colors corresponding to clusters specified in a external imported classification.
In the case ArrayMiner is run from GeneSpringä with the "ArrayMiner on Classification" option in GeneSpring's Navigator, the classification sent by GeneSpring is automatically loaded as a color scheme.

Macintosh version

Thanks to Optimal Design's proprietary library of portable graphic code, the whole of the graphical interface was ported to the Macintosh platform (OSX). Both Win32 and Mac OSX versions will be available simultaneously from this point on.

Clustering Algorithm

Improved efficiency of local optimization

The efficiency of the local optimization used in the algorithm was improved, resulting in a better quality/computation-time ratio.

User Interaction

Macintosh GeneSpring™ interface

Although GeneSpring currently only runs in the OS9 compatibility window on Macs, a data exchange interface is now available between GeneSpring and ArrayMiner, the latter running solely under OSX.

Unclassified items supported in imported classifications

The import of external classifications now also supports unclassified items.

Graphical Interface
- New main graphic window
- 3D cube controlled by mouse dragging
- Mouse wheel support
- New Closest Clusters window
Clustering Algorithm
- Full support of missing values
- Up to 200000 genes supported
User Interaction
- Copy of cluster contents into the clipboard

Graphical Interface

New main graphic window

The main graphic window has been completely redesigned to centralize the most relevant information into one window. The window now joins the heat map of all clusters together with the list of genes in a selected cluster and their profile curves. Both the heat map and the profile representation panels are resizable. Selecting a cluster or a gene in a cluster is thus even easier than before.

3D cube controlled by mouse dragging

The 3D view of the data was enhanced by adding the possibility of controlling the orientation of the 3D cube by a simple mouse drag. The data are hence more easily viewed from any angle in the 3D cube, giving the user an even better "feeling" of the relationship between the various clusters and genes. Even heavy data are easily manipulated thanks to wire-frame representation of the cluster hulls during the orientation of the cube.

Mouse wheel support

The wheel of the mouse, when present, is now taken advantage of, for easy scrolling of lists, zooming in or out the 3D cube, and selecting the color scale of the heat map.

New Closest Clusters window

The window showing the clusters that are closest to a given cluster has been redesigned for better readability. The window is now resizable to offer the most appropriate viewing size, and it is possible to select a gene in the window by clicking its profile.

User Interaction

Copy of cluster contents into the clipboard

The list of genes, together with information pertaining to the fit within its cluster, is now easily copied from ArrayMiner to the clipboard, from where it can be pasted directly into other applications, such as Excel, for further processing outside ArrayMiner.

Clustering Algorithm

Full support of missing values

Missing values are now fully supported, under all distance measures (i.e., Euclidean, Pearson coefficient, etc.).

Up to 200000 ORFs supported

The maximum number of genes, previously limited to 35000, has been increased to 200000.

Graphical Interface and User Interaction
- Classification Compare tool

Graphical Interface and User Interaction

Classification Compare tool

The Classification Compare tool allows the user to compare different classifications, such as classifications obtained by clustering into different number of clusters, in a very intuitive way. This conveys an additional insight of the user's data and, in particular, helps estimate the number of clusters.

Graphical Interface
- Easier axis selection in 2D & 3D
- Gene list in main window
- Selection by clicking a profile in Activities window
- 3D Hull cluster selectable
- Gene search available in all views
- Isolate gene in 3D
User Interaction
- Help texts available on-line

Graphical Interface

Easier axis selection in 2D & 3D

The axes used in the 2D and 3D representation of the profiles are now easier to select, with their attribute names directly visible in that window.

Gene list in main window

The main graphical window now shows the list of genes in the selected group, listing their names and descriptions together with information useful for assessment of the quality of fit of each gene within its cluster. The list is can be sorted up or down by any of the columns.

Selection by clicking a profile in Activities window

In the Activities window, the user may now select a gene by clicking on the line representing its profile. This provides a useful connection between the graphical representation of a profile and the corresponding gene.

3D Hull cluster selectable

A cluster can now be selected by clicking its 3D Hull representation (previously, a point in the group had to be clicked). In the select mode for example, this makes it easy to hide a cluster when in the 3D Hull representation.

Gene search available in all views

In the main graphic window, the gene Search feature is now available in all representations (2D, 3D, xD).

Isolate gene in 3D

A new popup option available when right-clicking on a line in the gene list in the main graphical window (xD view) allows to "isolate" the gene in the 3D view. The representation switches to 3D view, and only two clusters are shown: the gene's own cluster, and the cluster which is closest to the gene (in expression profile distance terms) among the clusters which do not contain the gene. This allows for an easy assessment of the reasons why a gene was placed in its cluster rather than the "second best" other cluster.

User Interaction

Help texts available on-line

It is no longer necessary to press F1 for help on an item when specifying the clustering parameters, the help text is now available on-line in the bottom of the form.

Graphical Interface
- Convex hull option in 3D view
- Resize Window
User Interaction
- Wizard functionality in the main window
- Context-dependent Help on parameters
- Optimization Finished indication
Clustering Algorithm
- Log-transform of activity data
- Substantial RAM savings on large data

Graphical Interface

Convex hull option in 3D view

The popular three-dimensional view of the data is now even more appealing with the option of showing convex 3D hulls of the clusters in place of the points making up the clusters. This is especially useful on data featuring a large number of profiles. When all data points are displayed in the standard "points" fashion, the 3D cube may become cluttered by a large number of points, lessening the user's ability to grasp the relationships between the clusters. In the "3D Hulls" display, each cluster is represented by one three-dimensional convex hull, which makes is easier to asses the relationship between the clusters.

Resize Window

Each graphic window now features a "Resize window" option in its popup menu (accessible by right-clicking in the window). When activated, the option resizes the window to the largest size compatible with the image shown in the window, while also taking into account the constraint of the screen's size (i.e. windows do not span beyond the screen's boundary).

User Interaction

Wizard functionality in the main window

The Wizard panel has been replaced by the equivalent functionality within the main (text) window, yielding a more centralized command of ArrayMiner. As the user grows familiar with ArrayMiner, he or she may hide the explanatory text ("What is going on" and "What to do next"), to gain place on their desktop. Ultimately, the Wizard-like buttons themselves can be hidden for even more desktop space economy, ArrayMiner being then controlled by menus and keyboard shortcuts.

Context-dependent Help on parameters

Each of the various parameters that must be specified at the start of an ArrayMiner run now has an associated help text. Pressing F1 when filling in the parameters displays a short explanation of the meaning of the parameter being highlighted in the parameter list. Furthermore, there is additional help on values of some parameters. Thus for instance, there is an explanation of the meaning of the "Distance measure" parameter. In addition, when choosing the value for that parameter in the list of possible distance measures, there is an explanation of each of the possibilities (e.g. "Pearson Additive").

Optimization Finished indication

Although the optimization algorithm can be stopped at any time, ArrayMiner now features an indication on when it would best to do so. When the standard termination condition (i.e., no improvement since a given number of iterations of the algorithm) is met, a message informs the user that the results can be safely returned to GeneSpringä. The algorithm continues in the background though, allowing for a more thorough search when time is available.

Clustering Algorithm

Log-transform of activity data

ArrayMiner now gives you the possibility to log-transform the data before clustering, the option being available as a new parameter to set before optimization. When activated, the option replaces all activity data by their logarithm before any further processing (such as scaling for Pearson coefficient). This is recommended when the activity levels represent ratios, as a double increase of activity and a half-reduction of activity are then perceived as equivalent changes differing only by their sign.

Substantial RAM savings on large data

ArrayMiner now claims a substantially reduced amount of RAM on large data, the saving being as large as a few hundreds of megabytes in some cases.