Clustering Exercises (Fri Nov 24 2006)
   ======================================
   Follow-up course of Microarray Data Analysis
   Nov 20-24 2006, PICB Shanghai
   Angela Becker, Peter Serocka

1. Final view onto the data from the Manila course:
   Normalised expression data from slide 1 (named slide205809) are
   supplied in /picb/course/41-Manila-Expr-Data as .mev files.
   Load them into TIGR MeV and compare in the main "heat map" view. 
   As the data a supposed be replicates, how would you judge the
   reproducability? Does it make sense to apply statistics to 
   measure significant differential expression?

2. Prepare Gain-of-function analysis on drosophila:
   Two time-series expression tables of 1000 genes (selected by
   highest occuring foldness) are provided in /picb/course/60-Drosophila/ .
   Load drosophila-gof.txt (TAB-delimited) into TIGR MeV
   Use "Utility" -> "Search" to visualise Angela's results by defining
   three "clusters" based on the following words found in the annotation:
   i.   Angela-GOF     (->assign color red)
   ii.  Angela-LOF     (->assign color blue)
   iii. Angela-HR      (->assign color yellow)

3. Hierarchical clustering
a. Perform hierarchical clustering (HCL) of the gene with default parameters.
   How do Angela's groups of candidates (colored in red, blue, yellow
   due to step 2) show up in the tree?
a. Perform hierarchical clustering with different linkage types.
   Compare the results to the tree from step b.: Are the candidate groups
   arrange differently?

4. K-Means clustering
a. Perform k-means clustering (KMC) first with default parameters, then
   with "Pearson correlation" as a metric. In which way do the
   expression graphs of individual clusters reflect the different
   ways ("metrics") of measuring the similarities of expression profiles?
b. Perform KMC with default parameters, but ask for 20 clusters.
   Find small clusters with candidate genes as well as non-candidate
   genes. Find out more (from Angela) on how candidate genes where
   discriminated from non-candidates.

5. Other clustering methods
   Try out self-organized trees (SOTA) and self-organized maps (SOM),
   starting with the default parameters. 

6. Gain-of-function vs. Loss-of-function
   Load drosophila-lof.txt into a NEW, separate MeV array view
   and color-label annotated candidates as in step 2.
   Using the clustering method(s) of you choice, find
   non-candidates that are similar to the candidates and
   present your results to Angela.
   
7. Does it seem useful to load both GOF and LOF tables into a 
   common table and perform clustering?