• Cluster yeast cell cycle data using Average Linkage.
• Compare clustering results of Average Linkage, K-means and SPC
   

The programs Cluster and TreeView that are used in this exercise were written by Michael Q. Eisen (eisen@genome.stanford.edu)
and can be downloaded upon request from http://rana.stanford.edu/software/

The software package for SPC (Super-Paramagnetic Clustering) can be obtained by mailing to Prof. Eytan Domany (fedomany@wicc.weizmann.ac.il).

Goal: Find groups of genes with similar temporal behavior.

• Run Cluster
  • press Start (bottom left corner), select Programs, select Clustering, select Cluster.
  • NOTE: you can read the manual by pressing "Read  Manual".
• Load the yeast cell cycle data of an alpha factor arrest and release experiment. The data is a result
  of a cDNA chip experiment of the yeast cell cycle (taken from Eisen et al. (1998) PNAS 95-14863).
  In the file for each gene there is the log₂(expression at time i / expression in mixture of  asynchronous cells). 
  A missing value means that there was no strong enough signal in the spot.
  In the experiment measurements were taken over a period of 119 minutes which is nearly two cell cycles.
  • Press "Load File"
  • Load the file from C:\clustering\exercise\alpha.txt. (A copy of the file can also be found here).
  • Question: How many genes and how many time points were measured?
• Preprocess the data (center and normalize genes)
  • There is no need to filter the data.
  • Choose the "Adjust Data" tab: check the normalize genes and mean center genes.
  • Press Apply

• Cluster the genes using Average Linkage
  • Choose the "Hierarchical Clustering" tab
  • Since we want to cluster only the genes and not the arrays (time points), uncheck the Arrays cluster.
  • Choose "Correlation (centered)" as similarity measure.
  • Question: What will be the difference if you use "Absolute Correlation (centered)" as similarity measure?
  • Press Average Linkage Clustering. This creates two files alpha.gtr (dendrogram of genes) and alpha.cdt (reordered data matrix).
• View the clustering results in TreeView
  • run TreeView by pressing Start (bottom left corner), select Programs, select Clustering, select TreeView.
  • Load the file C:\clsutering\exercise\alpha.cdt. Use the menu option File and then Load.
  • Scroll along the tree. Note: green color means negative normalized log ratio, red color means positive.
• Search for genes which are known to be cell cycle regulated: HHT1 (S phase), CLN1 (Late G1), CLB1(G2/M)
  • Choose Find in the menu, then Gene, and type HHT1. Press Find.
  • Select the cluster of genes that contains HHT1 by clicking on a vertex on the dendrogram or
    by using up/left/right buttons to go up the tree/down to left descendant/down to right descendant.
  • Question: How many genes did you choose?
  • Question: Are the genes in the cluster from the same biological process?
  • Save the data of this cluster by selecting File/Save Data. Change the name to hht1.txt.
  • Open Excel (Start/Program/Microsoft Excel).
  • Load hht1.txt. (File/Open C:\clustering\exercise\hht1.txt). Press Finish in the Text Import Wizard.
  • Create a graph of the normalized gene expression over time.
    • Select the region C2:T9 by dragging the mouse from C2 to T9 while pressing the left mouse button.
    • Select the Insert/Chart. Line. Choose the first sub-type. Press Next. Make sure the rows button is chosen.
      Then, press Finish.
    • Question: At what times does the expression of these genes peak?
    • Save the Excel work book by selecting File/Save and changing the Save as Type to Microsoft Excel Workbook.
      The filename should be hht1.xls (result C:\clustering\exercise\hht1.xls).
  • Repeat this analysis with genes CLN1 and CLB1 (C:\clustering\exercise\cln1.xls, C:\clustering\exercise\clb1.xls).
  • Question: Order the phases along the time using the peak times you recorded.
  • You can look at additional interesting clusters. Look for gene UBP6, SPA2.

Goal: See the differences between the methods.
K-Means

• Cluster alpha.txt as before using K-means with K=150.
  • Return to Cluster. If you closed it, rerun it and normalize the data as in part A.
  • Choose the K-Means clustering.
  • Check only "Organize Genes".
  • Choose K to be 150. 150 will give an average cluster size of ~16 (2467/150) genes.
  • Leave max cycles 100.
  • Press Execute. It could take a while. Follow the iterations at the bottom of the window.
• Load the resulting file in Excel (C:\clustering\exercise\alpha_K_G150.txt).
• Find the cluster that contains CLN1. The clusters in this file are separated by empty lines that have NONE in column A.
• Plot the normalized gene expression of the genes in the cluster.
• Compare your results to part A.

Super-Paramagnetic Clustering (SPC)

• Load the SPC results in TreeView (C:\clustering\exercise\alpha_SPC.cdt; or get files from alpha_spc.cdt, alpha_spc.gtr).
• Find the cluster that contains CLN1.
• Question: Is it easier to decide which genes belong to the cluster?
• Save the data for this cluster in C:\clustering\exercise\spc_cln1.txt.
• Load the file in Excel. 
• Plot  the normalized gene expression of the genes in the cluster.
• Question: Compare your results with to K-means and Average Linkage.
• Question: Draw a Venn diagram that counts the number of common genes in all methods.
  

Back to the Workshop Home Page