Notes 20100914.143119 BIOL 614 McConkey Bioinformatics Course

From SnOwy - Ed's Wiki Notebook

Introduction

• Students' knowledge assumed to be: knows how to use BLAST and understand its algorithms; may know tBLAST (translated blast); not likely to know about BLAST's e-values.
• This is a practical course -- theory is learned up to the point where we understand why we would select a particular tool
  • this course will be good to fill in the knowledge gaps

Topics

• Sequence alignment and BLAST statistics
• MSA -- 'a little more information'
• Clustering, phylogeny
  • Clustering: describes taxa that are similar
  • Phylogeny: see Müller's course for more.
• protein structural modeling
  • binding sites, predictions etc.,
• functional annotation: genes, proteins
• data analysis in genomics, proteomics
  • metagenomics etc.
• microbial genome sequence assembly

Topics from the floor

• next gen sequencing
• CNVs
• Perhaps: machine learning methods

Statistics

• various numbers to detect errors, inaccuracies
• multiple hypothesis testing etc.,

Evaluation

Independent Project

• projects should be related to research
  • impetus of course is to use discussed methods with respect to own academic interests

Seminar Component

• discuss the project selected
• includes also tool theory
• roughly 30 minute presentation

New in this version

• Lecture component
• Lab component
  • C2-160: Friesen Computer Lab - Thursdays 2:30pm to 4:30pm.
  • designed as a set of example cases
  • gives practical understanding of what to expect in terms of returned data from tools

Notes

Scoring Matrices and BLAST Scoring

• Read "What is dynamic programming" - Sean R Eddy
• Read "... BLOSUM62 ..." - Sean R Eddy

Pairwise alignment

• try to pick out the best scoring alignments
  • depends on objective function
  • should also estimate that the alignment occurred just by chance (E-values)
• scoring functions depend on amino acid or nucleotide mutation
• Expectation value: the probability of an alignment occurring by chance alone
• low-complexity regions
  • sequences that contain very short repeated sequences
  • low complexity filters: removes tendency to align such regions
  • example: leucine rich regions that form alpha helices.
• align DNA if no protein coding region available
• multiple domains
• entire domain insertions or deletions
• Score -- raw scores, bit scores
  • Raw score is the raw alignment score
  • Bit score is derived from the algorithm used
  • Both scores can be used to derive some e-value

Deriving Scoring Matrices

• let q_a,b be probability that an alignment of amino acids a, b occur in homologous sequence pair
• let p_a, p_b be overall frequencies of occurrence of a, b
• probability of random alignment is the product of the aa frequencies p_ap_b
• log odds ratio for a, b is ...
• Odds ratio = q_a,b / p_ap_b
• Odds ratio === p(alignment is authentic) / p(alignment is random)
• we assume that each column is an independent event...
• total odds ratio = Π_u(q_a,b / p_ap_b)_u
• logarithm allows us to sum instead of multiply -- computationally and numerically more sane.
• S = Σ_ulog(q_a,b / p_ap_b)_u = Σ_u(S_a,b)_u

BLOSUM

• based on local alignments
• developed after PAM matrices -- more data
• derived from BLOCKS database
• BLOCKS Substitution Matrices = "BLOSUM" -- Henikoff and Henikoff 1991, 1992
• differences with PAM matrices:
  • About PAM matrices
  • PAM matrices attempt to fit an evolutionary model -- i.e. has the property that multiple applications yields a more evolutionarily distant pair of sequences
  • Used to identify conserved regions.
• Open question?! Transitive closure?!
• Why BLOSUM62? -- seemed to work the best.

BLAST

• different BLAST programs specialize for different kinds of tasks.

PAM

• PAM1 → highly similar
• PAM250 → highly divergent