Visualization is an integral part of avadis™ and all analysis results have a graphical output associated with them. Multiple graphical visualizations of data and analysis results help uncover patterns in the data. avadis™ offers powerful visualizations that are interactive and allow dynamic changes. This provides an easy option of visually mining the data and reducing its dimensionality.

The data views provided are the Spreadsheet, the Scatter Plot, the 3D Scatter Plot, Profile Plot, Heat Map, Histogram, Profile Match, Spot Image, Annotation View, Matrix, Summary Statistics, and Bar Chart.

All the active views are lassoed, i.e., selections on one view are propagated to all other open views. All views can be customized for publication quality images.

[ Click to Enlarge ]

[ Click to Enlarge ]

• each project has an associated project-type
• each project-type has a specific predefined Workflow
• several project-types come prepackaged with avadis™
• each workflow is a set of funtions organized logically into simple URL-type links in the workflow browser
• data, all its derived subsets and data views are organized into a hierarchical tree structure in the Navigator pane

• all functions in avadis™ are scriptable into workflows.
• any complex custom workflows, like SOPs, can be automated as custom Workflows.

The Affymetrix project in avadis™ has a highly evolved workflow equipped with latest probe summarization algorithms viz. PLIER, RMA, gcRMA, MAS5. It is in tune with the latest in GeneChip® technology.

For 1-dye and 2-dye analysis, all templates are available in the wizards to import data from all leading array platform technologies. These wizards can import data from all databases viz. Oracle, MySql, Postgres.

[ Click to Enlarge ]

[ Click to Enlarge ]

The avadis™ main window is divided into three main areas: the Desktop, the Navigator and the Workflow Browser.

The Desktop accommodates all the views and algorithm results pertaining to each project, the Navigator displays all currently open datasets, views and algorithm result reports in a hierarchical tree structure, while the workflow browser allows application specific workflows to appear as a sequence of user clickable links.

[ Click to Enlarge ]

[ Click to Enlarge ]

One of best achievements of avadis^TM is the fact that it has taken the most painful part of data analysis and packaged it into a handful of simple one-click functions.

Normalization within and across arrays is an important step in the analysis pipeline. Data for different arrays or from different dyes on the same array cannot be compared directly due to several sources of variation across arrays and dyes, e.g. experimental conditions, dye efficiencies, ambient light intensities etc. Normalization is a process by which all sources of variation that are not due to the experimental conditions are factored out. avadis^TM offers several normalization methods for both single dye and two-dye arrays along with visualization tools for determining the type of normalization needed, if any, and for assessing the efficacy of various methods. All normalization algorithms are based on the premise that most genes are not differentially expressed across arrays or channels.

avadis^TM offers three kinds of normalization:

• Mean/Median Shifting
• Linear and Non-Linear Lowess Regression and
• Quantile Normalization for Replicates


avadis^TM provides a variety of statistical tests like paired/unpaired T-tests, Mann Whitney test, n-way ANOVA and Repeated measures test for replicate experiment designs. The other statistical analysis options in avadis^TM; are Multiple group analysis, P-Value generation via permutations or asymptotic analysis, Multiple testing correction and interactive filtering based on p-Value, fold-change and the number of genes expected by chance.

avadis^TM supports a fast PCA implementation along with an interactive 2D viewer for the projected points in the smaller dimensional space. The scatter plots of data are projected along the principle components. It clearly brings out the separation between different groups of rows or columns whenever such separations exist.

avadis^TM performs PCA on a 20,000 x 1,000 dataset in under 5 seconds even on a 256 MB RAM desktop computer.

[ Click to Enlarge ]

Cluster analysis is a powerful way to organize rows in the dataset into groups or clusters of similar rows. There are several ways of defining the similarity measure, distance between two rows. While some methods are purely mathematical, others use domain specific knowledge about the rows.

avadis™ clustering module offers clustering algorithms like k-Means, Hierarchical, EigenValue, Self Organizing Maps (SOM), Random Walk and Principal Components Analysis (PCA) along with variety of distance functions like, Euclidean, Square Euclidean, Manhattan, Chebychev, Differential, Pearson Absolute and Pearson Centered.

Since different algorithms work well on different kinds of data, this large option of algorithms and distance measures ensures that a wide variety of data can be clustered effectively.

All clustering results are interactive views such as the clusterSet View, the Dendrogram View and the Similarity Image View. These views allow drilling down into subsets of data and collecting together individual rows or groups of rows, which look interesting into new datasets for further analysis. All views are lassoed, and enable visualization of a cluster in multiple forms.

[ Click to Enlarge ]

Classification algorithms are a set of powerful tools that allow researchers to exploit microarray data for learning-based prediction of outcomes of gene expression. These tools stretch the use of microarray technology into the arena of diagnostics and understanding the genetic basis of complex diseases.

Typically, classification algorithms can be applied to microarray data in two ways. The first type works at the level of individual genes. For example, if expression profiles as well as function information are available for a collection of genes, then this information can be used to learn a model, which can then predict functions for new genes given their expression profiles alone. The second type works at the level of experiments or samples. For example, given gene expression data for different kinds of cancer samples, a model, which can predict the cancer type for a new sample, can be learnt from this data.

The models built by these algorithms range from visually intuitive to very abstract. Further, the classification algorithms vary in their ability to handle multiple classes and discrete variables (only DT can handle discrete variables, e.g., tumor samples may be marked as large, small or medium and this may be one of the factors in learning a model). Together, these methods constitute a comprehensive toolset for learning, classification and prediction.

avadis™ has four powerful machine learning algorithms like Decision Tree (DT), Neural Network (NN) and Naïve Bayesian (NB) to classify samples or genes into discrete classes. In addition, a Linear Multivariate Regression algorithm allows for prediction of continuous variables like survival indices.

avadis™ classification module comprises a set of supervised learning algorithms, which construct a model from a training dataset. This model is then used to predict classes for new unclassified data.

[ Click to Enlarge ]

[ Click to Enlarge ]

Scripting enables automating repetitive workflows. Most of the functionalities in avadis™ are scriptable. Scripting is done using Python. Extensive help is available, including sample codes for all functions to help users with scripting.

The Log View maintains an audit trail of all analysis performed and modifications done to the dataset by the user.

Lasso View window shows all points selected in any view

Filter is a visual interface to dynamically filter views based on various combination of parameters

[ Click to Enlarge ]

[ Click to Enlarge ]

Typically, an organization will have several legacy data sources spread over multiple locations. These data sources often have diverse structures and access privileges. avadis™ enterprise provides the solution to these organizations based on a Client-Server architecture.

Data Integration - avadis™ enterprise features a client-server architecture in which access and integration routines can be set up at the server and automatically inherited by all clients across the entire organization. There are appropriate tools at the server and client ends for setting up links and forms for access to various data sources.

Collaborative Analysis - End-Users can save their analysis results and reports at the server and make these accessible to their collaborators, again just by clicking on links and filling up forms. The entire current session on one user’s client can be replicated on another user’s client for quick collaborative analysis.

Role Based Usage & Customization - avadis^TM enterprise enables role-based usage via access control for each data access link, form, or workflow and allows IT personnel within an organization to quickly customize and manage integration modules and workflows specific to the organization.

[ Click to Enlarge ]