Notes 20101012 BIOL 614

From SnOwy - Ed's Wiki Notebook

Dr. McConkey's bioinformatics tools class

Contents 1 Phylogenetic Models 2 Evolutionary Distance 3 Evolutionary Models 3.1 Nucleotides Evolution 3.2 Protein Evolution 4 Confidence measures

Phylogenetic Models

• model evaluation is useful for testing the applicability and "performance" (given particular information criteria)
  • see PhyML
• notice the difference between BLOSUM62 and LG (newer substitution model)
  • the first is based on probability of changing from one amino acid to another
  • the second provides a rate of change to allow us to build up a branch length
    • I think that the notion of "rate" is the same as the notion of "likelihood"
• note that for model assessment, the n-parameters term refers to independent parameters
  • in the set {π_A, π_C, π_G, π_T} -- the abundance of each nucleotide -- counts as only one parameter (they also always sum to one).

Evolutionary Distance

• simplest model: fractional alignment distance (p-distance).
• Poisson distance: d_p
  • example: PAM -- transitive application of mutations as rates for single aligned columns
  • p_1/2 = 1 - e ^e-rt
  • p = 1 - e ^e-2rt
  • ... ugh -- missed the last two equations that explain how to get to d_p -- distance
• Gamma distribution
  • models variation across different sequence positions
  • may be important for my project -- that is, mutation rates, highly conserved regions --
  • parameter a is inversely related to conservation
  • this is not a linear relationship

Evolutionary Models

In reality, for most biology related papers, only a single line of explanation about model selection is needed -- the model selection is powered by perhaps ProtTest, PhyML or some other automated model fitting software.

Nucleotides Evolution

• Jukes-Cantor -- all SNPs have the same probability
• Kimura -- added transitions and transversion scores
• HKY85 -- added relative abundance of {G, C}, {A, T}
• up to modern day -- GTR -- general time reversible -- the rates for each transition and for each transversion are different. Wow.

Protein Evolution

• PAM
• BLOSUM
• JTT (Jones, Taylor, Thornton) -- an extension and update of the original PAM matrix

Confidence measures

• bootstrap confidence
  • is Monte Carlo sampling -- each aligned column is pulled as a sample -- sample with replacement -- 1000 samples are the norm