Introduction to Bioinformatics 20120

1
Introduction to Bioinformatics20120

• Gianluca Pollastri
• office CS A1.07
• email gianluca.pollastri_at_ucd.ie

2
Credits

• Richard Lathrop and Pierre Baldis Bioinformatics
  courses at University of California _at_ Irvine.

3
Course overview

• Context DNA, RNA, proteins
• Resources GenBank, PDB, etc.
• Algorithms for sequence comparison.
• Phylogenetics.
• Structural bioinformatics protein structure
  prediction.

4
Lecture notes

• http//gruyere.ucd.ie/2007_courses/20120/
• confidential..

5
Recommended/useful readings

• No book is actually required
• Introduction to Bioinformatics
• Lesk
• Introduction to Computational Molecular Biology
• Setubal, Meidanis
• Bioinformatics the Machine Learning approach
• Baldi, Brunak

6

• CS 20120, Introduction to Bioinformatics
• Assignment 1, 29 January 2007
• 10 of the overall mark
• To hand in by midnight of February 12
• 1. identify your favourite pet
• 2. get the protein sequence for one of its genes
  on
• a. http//www.ncbi.nlm.nih.gov/entrez/
• 3. BLAST your sequence against UniProt at
• a. http//www.ebi.ac.uk/blast2/index.html?UniProt
• 4. If you get less than 6 results from 6
  different organisms, go back to 2 and choose
  another protein
• 5. Select 6 sequences returned by BLAST, from 6
  different organisms (ticking the appropriate
  boxes and downloading them in fasta format will
  give you the right input format for the next
  step)
• 6. Run clustalW on them using the page (be
  patient, might take time)
• a. http//www.ebi.ac.uk/clustalw/index.html
• 7. Draw a phylogenetic tree for your guide tree
  (.dnd) using an online viewer, e.g.
• a. http//bioweb.pasteur.fr/seqanal/interfaces/dra
  wtree.html
• 8. email me (gianluca.pollastri_at_ucd.ie)
• a. your protein sequence UniProt record

7
Algorithms for sequence comparison

• Generating all possible alignments and picking
  the best one impossibly slow.
• Dynamic programming (here programming has
  nothing to do with computers) solving a problem
  by splitting it dynamically into subparts.
• We build up a solution based on similarity
  between prefixes of the two sequences..

8
Aligning prefixes

• Specifically, we solve the alignment problem of
  two sequences by splitting it iteratively (or
  recursively) into the alignment of their prefixes.

9
the algorithm

• We can fill an (n1)x(m1) matrix with this
  stuff

10
Hope its right
11
Computing the matrix

• m s
• n t
• for i0..m ai,0 ig //m1
• for j0..n a0,j jg //n1
• for i1..m
• for j1..n
• ai,j max(ai-1,jg,
• ai,j-1g,
• ai-1,j-1p(si,tj))
• // (n1)(m1) max3 sums, etc.

12
Computing the alignment

• What we computed here is the max similarity
  matrix between all prefixes of s and t.
• Using this matrix we can compute the optimal
  alignments between s and t (they could be more
  than one).
• am,n is the max similarity between s and t. We
  find the alignment by tracing the choices that
  led us there.

13
Computing the alignment (2)

• // We have filled in matrix ai,j before
• ns
• mt
• al_s //store here aligned s
• al_t //store here aligned t
• gap2 //gap penalty
• inm //index for the alignment don't know how
  long, but at most nm
• align()
• while (ngt0 mgt0)
• if (ngt0 an,man-1,m-gap)
• al_sisn
• al_ti'-'
• nn-1
• else if (ngt0 mgt0 an,man-1,m-1-p(sn
  ,tm))
• al_sisn
• al_titm
• nn-1

14
align() while (ngt0 mgt0) if (ngt0
an,man-1,m-gap) al_sisn
al_ti'-' nn-1 else if (ngt0
mgt0 an,man-1,m-1-p(sn,tm))
al_sisn al_titm nn-1
mm-1 else if (mgt0 an,man,m-1-gap
) al_si'-' al_titm
mm-1 ii-1
15
(No Transcript)
16
Alignment

• ACC-AGGCTACGA
• ACCTGGGCCACGT
• only one gap, no big deal here..

17
Order matters

• There might be multiple paths with the same
  score. We used an upmost order here

1 vertical 2 diagonal 3 horizontal
18
Order matters (2)

• To follow a downmost order, reverse the if
  statements in the code.
• 1 horizontal
• 2 diagonal
• 3 vertical

19
(No Transcript)
20
Upmost and downmost alignment

• upmost
• ACC-AGGCTACGA
• ACCTGGGCCACGT
• ---
• downmost
• ACCAGG-CTACGA
• ACCTGGGCCACGT
• ---

• Both alignments have the same score
• 9 matches (1 x 9),
• 3 mismatches (-1 x 3),
• 1 gap (-2)
• 4

21
NW algorithm issues

• Always looks for a global alignment.
• If I try to align the following
• first_alignment_try_ok
• second_alignment_try_ehm
• This is what I get
• -first_alignment_try_-ok
• second_alignment_try_ehm
• -------

22
NW algorithm issues

• Always looks for a global alignment.
• If I try to align the following
• alignment_alignment_try_ok
• alignment_try_ehm
• This is what I might get
• alignment_alignment_try_-ok
• alig----------nment_try_ehm
• --

23
Local alignment

• We may want a variation of the previous algorithm
  that throws away stuff that clearly does not
  match, while keeping the good bits, together.
• More formally find the highest scoring alignment
  between substrings of s and t.

24
Smith-Waterman algorithm
Mike Waterman at a conference