Notes 20101005 BIOL 614 McConkey Class

From SnOwy - Ed's Wiki Notebook

Last week: HMM (HMMER) / PSSM (PSI-BLAST)

Next week: Models of evolution -- backend scoring functions

Contents 1 Topics for today 2 Phylogeny 2.1 Molecular phylogeny overview 2.1.1 Selecting sequences 2.1.2 Sequence data -- DNA / RNA / Protein 3 Multiple Sequence Alignments 3.1 ClustalW 3.2 Expectation Values Can't Be Used Here 3.3 Conserved Domain Database 4 Phylogenetic Trees Continued 4.1 MSA Continued 4.2 MSA - Features 4.3 Mr. Bayes - Software 4.4 Parsimony vs Likelihood 4.5 Evaluation of method 4.6 Bootstrapping -- robustness checking 4.7 Human friendly displays 4.8 Species vs gene, protein trees 4.9 Additive vs. non-additive trees

Topics for today

• multiple Sequence Alignments
• phylogenetic Trees (intro)

Phylogeny

• we can consider classical taxonomy as phylogenetics -- we now take molecular data instead of phenotypical characters
• ~100 years old
• requires: similarity score
• ATGC - A simple, fast, and accurate algorithm to estimate large phylogenies by maximum likelihood.
  • is a web server program
  • input: PHYLIP / MSA -- DNA / AA / Interleaved / Sequential
  • substitution model: scoring functions
  • transition / transversion ratio
  • substitution rate categories
  • gamma shape parameter

Molecular phylogeny overview

• in this class: molecular data
• data analysis
  1. selection of sequences
  2. multiple sequence alignment
     • substitution model (BLOSUM, PAM)
     • handling indels
       • possibly delete entire column?
  3. tree building
  4. tree evaluation
     • why are we building a tree in the first place?
     • xenologs (horizontal transfer) -- paralogs (gene duplication) -- orthologs (evolution)

Selecting sequences

• figuring out the phylogeny of those organisms
• CoG -- clusters of orthologous groups

Sequence data -- DNA / RNA / Protein

• protein works better in phylogeny for distantly related organisms
• DNA, RNA works better for closely related organisms
• able to use ribosomal RNA sequences -- this is an exception, this is present in many organisms
• fine details -- protein sequences are going to be too similar
  • DNA sequences have more changes for each sequence
  • DNA translated into proteins sequences to create primary alignment -- DNA however is used finally
• DNA synonymous substitutions can be detected this way
• single nucleotide substitutions, sequential (temporally, not spatially) substitutions, coincidental substitutions
• DNA is closer linked to the molecular clock
• DNA -- noncoding regions, pseudogenes, transversion, transition rate

Multiple Sequence Alignments

• building a MSA (ClustalW)
  • progressive alignment - sequences aligned in a specified order
• local / global weighting (T-Coffee)
  • information from other sequences in pairwise alignment
  • similar regions are downweighted ...
• ungapped segment alignment (DiAlign)
  • no gap penalty; aligns short global regions, merges sequences (with gaps)
• iterative re-alignment (MUSCLE)
  • sequence subsets removed and re-aligned

ClustalW

• start of with n sequences
• perform a (complete) set of pairwise global alignments -- n(n-1) ÷ 2 alignments
• guide tree created -- most similar to least similar
• progressive align sequences -- dynamic programming

Expectation Values Can't Be Used Here

• moot when it comes to MSA
• we've already assumed that we're aligning homologous sequences
• we don't really ever calculate e-values for that reason
• scoring functions thus assess something else...

Conserved Domain Database

• CDD will also return a functional summary

Phylogenetic Trees Continued

MSA Continued

• an initial error can occur for the primary pairwise alignments

MSA - Features

• conserved regions -- hydrophobic transmembraneous
• cysteine (disulphide bridges) should be conserved

Mr. Bayes - Software

• will mask away a certain region riddled with gaps to compare available information
  • that is, will treat sequences as though they have the same length in primary tree construction
• why am I interested?
  • is this the "view" concept in my first subproject?

Parsimony vs Likelihood

• parsimony: fewest changes (shortest edit distance)
• maximum likelihood: parameterizes the evolutionary functions involved and provides lowest scoring path
  • but does likelihood actually have a good foundation in reality?
  • we have difficulty observing point mutations let alone bigger insertions and nucleotide repeat diseases ...

Evaluation of method

• consistency -- given different information (trials, genes, columns) within a single dataset
• robustness -- given a injected random mutations
• efficiency -- outcome (accuracy, precision, conciseness) ÷ price (time, computation, complexity)

Bootstrapping -- robustness checking

• can we keep finding the same branching order if we randomly permute a subset of information from the same dataset (columns are multiple trials)
• scales linearly with the number of sequences input (makes sense, reperforms the experiment that many times)
• the MSA -- sans columns with gaps -- the trials get shuffled

Human friendly displays

• cladogram -- phylogenetic branch lengths are not to scale -- pretty for a diagram
• phylogram -- phylogenetic branch lengths are to scale

Species vs gene, protein trees

• species -- one representation that is orthologous for all members of the tree -- is the evolution for the entire tree
• speciation -- reproductive isolation
  • internal nodes represent species
• gene duplications, paralogues etc.,
  • allowed to occur in the tree -- is likely to be different from the species tree
  • gene families
• gene loss can change the drawn phylogenetic tree: missing taxon.

Additive vs. non-additive trees

• additive: preserves sum of distances between terminal elements
• additivity depends on dataset!
• phylogenetic trees should be additive
• non-additive: microarray data -- no particular assumption of common ancestor between clustered elements