Microarrays data analysis using Java visualization

1
Microarrays data analysis using Java visualization

• By Rui Ding
• ------University of Minnesota, Morris

2
Background

• Cancer classification is a very important
  research areas.
• Class prediction and class discovery are the goal
  of cancer classification.
• Use gene expression and make DNA microarray data
  to do cancer classification.
• Two common used Microarray.
• Statistical methods used in Microarry data.
• Visualization techniques used in Mircroarray data
• Our dataset is applied to human acute leukemias
  data. 7128genes for each cases

3
Backgroundintroduction to visualization technique

• The term visualization refers to an approach to
  transforming scientific data into understandable
  figures or graphs
• At present, visualization has been widely used in
  many different areas
• Web-based visualization.
• Why we use visualization?
• Java and Java visualization

4
Backgroundour objective

• Perform a Bayesian variable selection
• Perform a principal component analysis
• Use Java visualization technique to visualize
  gene data for classification
• Create the Java visualization website

5
Bayesian probit model with variable selection

• We model the expression data using a binary
  probit model.
• Bayesian approaches to statistical inference
  impose prior distributions on the unknown
  parameters.
• We use a Bayesian variable selection to identify
  important genes. One of Bayesian variable
  selection is the stochastic search variable
  selection (SSVS) by utilizing mixture priors for
  the regression coeficients to perform the
  selection

6
Principle Component Analysis

• Principle Component Analysis (PCA) is a method
  that reduces data dimensions.
• The purpose of PCA in this research is to reduce
  7128 gene variables in 72 different people to a
  smaller number of principle component.

7
Principle Component AnalysisPrinciple Component

• Linear combination of observed variables
• Y1 a11(X1) a12(X2) a1p(Xp)
• Then, the ith principal component is
• Yi ai1(X1) ai2(X2) aip(Xp)
• we should decide how many components we need to
  keep
• we could extract either two principal components
  or three principal components
• The first component accounts for a maximal amount
  of total variance in the observed variable.
• The second component is the second most of the
  amount of total variance in the observed
  variable, meanwhile, it is uncorrelated with the
  first component.
• The remaining components follow the same rule as
  the second principal component.

8
Principle ComponentCovariance matrix and
Correlation matrix

• There are two important matrix used in the PCA,
  covariance matrix and correlation matrix.

9
Principle ComponentEigenvector and criterion in
solving PCA

• Principal components and eigenvector
• Kaiser criterion (eigenvalue-one criterion)
• Using a specified proportion of variance

10
Principle Component AnalysisImplementation of
human leukemia data

• Extract three coordinates of 3D plot for JAVA
  visualization from Human Acute Leukemia data set
• 7218 gene expressions (Test dataset and Train
  dataset) in three classes of leukemia 38 cases
  of B-cell acute lymphoblastic leukemia (ALL), 9
  cases of T-cell ALL and 25 cases of acute myeloid
  leukemia (AML).
• Gene expression levels were measured using
  Affymetrix high-density oligonucleotide arrays.
• We obtained three principal components for 72
  cases which are a linear combination of 7218 gene
  expressions variables with coefficients equal to
  the components of an eigenvector of the
  covariance matrix.
• All computations for principle component analysis
  were performed on a personal computer using
  SASAnalyst (version 9.2).

11
Java visualizationjava 3D

• Java 3D was developed by Sun company it is a
  convenient approach to program interface used for
  writing three-dimensional graphics applications
  and applets.
• High performance in Java 3D
• Application of Java 3D

12
Java visualizationScene organization

• 3D Universe At the top of the tree structure in
  Java 3D, it controls the real-time rendering of
  the whole scene.
• Scene Root it stores all scene data and does a
  real-time rendering
• View Port A class of universe. It is used to
  control the location of the view, set the
  projection model, set and display a substitution,
  etc. We use the orthogonal projection model to
  transform from 3D to 2D.
• Objects In our system, we have two types of
  objects atoms, represented as spheres, and
  coordinate axis, represented as cylinders.
• Lighting This category sets up
  environment/direction light in this system.

13
System overview

• (1) control module
• (2) data module
• (3) visualization module

14
Visualization Module

• The role of the visualization module is to be in
  charge of visualizing data in a webpage.
• We use applet in our implementation.
• Two dimensional raw data.
• Three dimensional raw data.
• User is able to zoom in /out, translate, or
  rotate the image in the display panel

15
Control Module

• Scheduling between web pages, data forwarding,
  handling errors and so on.
• Our system involves only four pages.
• Two pages are static
• One is the data visualization page
• The error handling page
• The applet introduced in the last section display
  the figure by requesting the data from the
  server.
• This creates communication problems between an
  applet and the server.
• We use HTTP to communicate

16
Data Module

• The task of this data module is to parse the data
  file. This system involves the existence leukemia
  data files.
• For atom structure data, we take into account the
  higher-level data structure.
• We use object-oriented method to parse the file.
  It lets the object achieve self-analysis, then
  the code is easier to maintain.
• We construct the class structure
• (1)The ConfigModel class corresponds to the whole
  point structure data file. This class uses the
  configElemList property to save each
  configuration file's information.
• (2) configElem represents a configuration
  information. Its attributes specify the code of
  this configuration section. pointNum represents
  number of point. Similarly, this class uses an
  array (pointModelList) to store the point model
  data as the current configuration.
• (3) PointModel class represents a point model. We
  have one point models in this system. Therefore,
  we have constructed this class as an abstract
  class. It uses an array (pointList) to store 3D
  location data of each point.
• (4) Point class represents the x, y, z coordinate
  in the space.

17
ResultBeysian Probit Model

• Using the Bayesian probit model with variable
  selection, we selected 6 out of 7128 genes.
• The probability of these six genes are 962th
  gene 0.56, 2118th gene 0.63, 2402th gene 1.00,
  3433th gene 1.00, 4993th gene 0.86, 5603th
  gene 0.41.
• These six genes did not help to classify two (AML
  and ALL) or three groups (AML, B-cell, and
  T-cell), but those genes can be important
  biomarker which may be used to see how well the
  body responds to a treatment for a disease or
  condition.

18
ResultPCA

• 3D figure shows two groups, AML(blue), and
  ALL(red).
• PCA method is good statistical technique for gene
  classification

19
Conclusion and Future Work

• Two statistical methods to do microarray data
  analysis
• Bayesian probit model with variable selection, we
  selected 6 genes out of 7128 genes which can be
  important biomarker.
• we performed principle component analysis with
  7128 genes.
• The results of the PCA are much better to
  classify two (AML and ALL) groups compared to the
  method by the Bayesian probit model with variable
  selection.
• Our future study will be how to improve the
  classification by the Bayesian probit model.

20
Question

• Please feel free to ask questions

21
Thanks for you time

• Thanks Jong-Min!
• Thanks Engin and Jon!
• Thanks all people that help me to complete the
  goal of seminar.
• End