The objective of this assignment is to familiarize you with basic gene expression profile classification analysis.
Data:
We work on the subtype classification of acute lymphoblastic leukemia (ALL) from Yeoh et al (Cancer Cell, 1:133--143, March 2002). A subset of the data from the paper is used. Please download the data from
http://www.comp.nus.edu.sg/~wongls/courses/cs2220/2015/Assignment2.zip.
Tool:
The WEKA machine learning package is used in this assignment. Please download and install it from http://www.cs.waikato.ac.nz/ml/weka/.
Q1 [3 marks]: The data are provided to you in individual files, one for each sample. The files are grouped into a training and a testing set.
How many training samples of each ALL subtype are there?
How many testing samples of each ALL subtype are there?
Do all the sample files have the same number of genes?
Q2 [3 marks]: The first step in any gene expression analysis is to clean up the data. A simple way to do this is to remove all genes that belong to one of the following categories: (i) Affymetrix control genes, (ii) genes that are absent or are marginal in at least one sample, (iii) genes that do not appear in all samples. Please ensure that exactly the same genes are removed from both training and testing samples. [You can do this by hand or by writing a computer program.]
How many genes are left after the cleaning above is performed?
Q3 [3 marks]:The cleaned samples from the training set should be merged into a single training file. Similarly the cleaned samples from the testing set should be merged into a single testing file. These two files should be in CSV or ARFF format that can be read by WEKA. [You can do this by hand or by writing a computer program.]
Please pass me a copy. If they load properly, you get 3 marks.
Q4 [3 marks]: Use the “chi-square” filter in WEKA to select the 200 most discriminative genes based on the training data.
List the genes that are selected and their chi-square values.
Q5 [5 marks]: The genes selected in Q4 may not be specifically discriminative for a specific subtype. Describe what you will need to do if you have to select the 30 most discriminative genes for each subtype.
Q6 [8 marks]: Build a SVM classifier and a C4.5 classifier on the training data using the 200 genes selected in Q4. Test these two classifiers on the testing set.
Show the confusion matrix for these two tests.
What is the accuracy of these two classifiers?
Q7 [5 marks]: Suggest a possible way to improve the accuracy of the classifiers built in Q6.
--------------------end of assignment #2--------------------
