Daniele Merico - HowTo Directory

Affymetrix Microarray Analysis

Importing raw data and generating standard gene expression metrics (signals, calls, etc...)

• [:DanieleMerico/HowtoDirectory/AffyCelCalSig: Importing Affymetrix CEL files, calculating MAS5 calls and signals]BR CEL files are the almost-raw files generated after chip image processing by Affymetrix software; BR the "fun" usually starts from the CEL files onwards; here's is the simplest things you can do with CEL files.

• [:DanieleMerico/HowtoDirectory/ExprSet: Importing Affymetrix CEL files, bothering about the R exprSet object, calculating MAS5 calls and signals]BR if the experimental design is quite complex, or you are using a function requiring an expression set (exprSet),BR then, sorry, but you probably need to read this part instead of the previous one.

Computing Differential Expression

• 2-class methods BR these methods require a dicotomic classification of the samples (e.g. case vs control), and reproducibility of samples belonging to the same class

  • [:DanieleMerico/HowtoDirectory/PLGEM: PLGEM]BR Features:

    • statistic used: corrected signal-to-noise, every gene treated as an independent entity; signal-to-noise is corrected according to an error model for the global estimation of varibility;
      • error model requires: linear relation between signal mean and standard deviation
    • significance: estimated by randomly permuting the data (by column), and computing the statistic;
    • recommended when: the number of replicates is uneven between case and control, with one of the two having very few, or just one replicate;

    • proteomics: successfully applied to tandem mass-spec proteomics data, where the signal was generated as abundancy-normalized peptide counts (NSAF)BR

    References
    • Pubmed.ID: 15606915 (main)
    • Pubmed.ID: 18029349 (proteomic application)

  • SAMBR Features:

    • statistic used: corrected signal-to-noise, every gene treated as an independent entity;
    • significance: estimated by randomly permuting the data (by column), and computing the statistic;
    • recommended when: the number of replicates is 3 or more, and even between case and control;

    • proteomics: unknownBR

    References:
    • Pubmed.ID: 11309499 (main)

General Computational Techniques

Computational Techniques for multi-dimensional data:

• [:DanieleMerico/HowtoDirectory/Distances: A few tips on distances] (especially for binary strings)