Introduction to Microarray analysis

From Organic Design wiki

{{#security:edit|Sven}} {{#security:*|Sven}}


Overview of experimental process

File:Expt.png

(Courtesy Mik Black)
  • Competitive hybridization to spotted oligo/cDNA transcripts
  • Interested in genes that change between treatment conditions
differential expression versus equivalent expression

Statistical analysis process

File:Process.png

  • Raw data (GPR file format)
http://www.moleculardevices.com/pages/software/gn_gpr_format_history.html
  • Each GPR intensity file is typically >8 megabytes
  • Each TIFF image file is typically >30 megabytes
  • A microarray experiment consists of several → many slides

Statistical issues

  • In the past statistics was developed for n >>p
n observations, p variables
  • Gene expression data n<<p
Thousands of measured genes (p)
Small number of biological replicate slides (n)
  • Gene expression data can be highly correlated
groups of genes are regulated in the same way
  • Data not normally distributed
log transform highly skewed intensity data

File:Graph channels.png


Analysis wish list

  • Ideally would like unambiguous interpretation of results
  • Large amounts of data to analyse can be overwhelming and make interpretation subjective
  • Independent reproducibility of results by another collegue
Keep a record (log file) of what was done

Analysis aim

  • Obtain a list of genes which we think are differentially expressing
      Block Row Column            ID   Name         M        A         t      P.Value        B
10396    20  15     23 171121_390_49 171121  5.035364 13.25087  49.62425 3.220044e-05 11.27486
4517      9  13      9  20264_118_53  20264  4.396719 11.11976  47.06004 3.220044e-05 11.05671
16881    32  21     22 165415_634_53 165415  4.645384 12.65872  43.40359 3.220044e-05 10.70650
16086    31  10      9 185903_436_49 185903  5.146504 11.36911  42.75724 3.220044e-05 10.63926
6508     13   7     22 197386_457_55 197386  4.621024 13.20426  42.09902 3.220044e-05 10.56899
5471     11   8     20 142178_355_53 142178  4.795734 12.07427  41.23346 3.220044e-05 10.47374
8395     16  20     23   251706_1_53 251706 -5.003475 13.04571 -38.61325 3.220044e-05 10.16421
4330      9   5      6 297409_340_47 297409  4.421922 12.27208  38.52215 3.220044e-05 10.15284
12479    24  14     13 163360_396_47 163360  4.367943 11.10478  38.21662 3.220044e-05 10.11439
15024    29  10      5 149243_674_53 149243  4.372419 11.36572  37.86362 3.220044e-05 10.06935
  • Easier to rank genes in order of evidence of differential expression than it is to select a specific cutoff
  • If we do select a cutoff, False Discovery Rate (FDR) cutoff is usually used
FDR threhold is the expected proportion of genes in a list that are likely to be incorrect