BLAST Michael Schroeder Biotechnology Center TU Dresden Contents - - PowerPoint PPT Presentation

Michael Schroeder

Biotechnology Center TU Dresden

BLAST

Contents

• Why to compare and align sequences?
• How to judge an alignment?
• Z-score, E-value, P-value, structure and function
• How to compare and align sequences?
• Levensthein distance, scoring schemes, longest common subsequence, global

and local alignment, substitution matrix,

• How to compute an alignment?
• Dynamic programming
• How to compute an alignment fast?
• BLAST
• How to align many sequences
• Multiple sequence alignment, phylogenetic trees
• Alignments and structure
• How to predict protein structure from protein sequence

2

Motivation

• Two sequences of length n
• Dynamic programming:
• matrix = time proportional to n2
• Database of m sequences:
• time proportional to m n2
• Manual search: How long does it take?
• 1 cell = 10 sec
• 1.000.000 sequences
• Sequence = 100 amino acids

3

Motivation

• How can we reduce the database size?
• How can we reduce the matrix size?

4

lev(petra, peter) = ?

5

Levenshtein Distance with Dynamic Programming:

Are all cells in the matrix equally important?

6

i \ j p e t e r 1 2 3 4 5 p 1 1 2 3 4 e 2 1 1 2 3 t 3 2 1 1 2 r 4 3 2 1 1 1 a 5 4 3 2 2 2

Levenshtein Distance with Dynamic Programming:

Are all cells in the matrix equally important?

7

i \ j p e t e r 1 2 3 4 5 p 1 1 2 3 4 e 2 1 1 2 3 t 3 2 1 1 2 r 4 3 2 1 1 1 a 5 4 3 2 2 2

For the two alignments, we only used 8 out of 36 cells. Can we discard the other cells beforehand? How?

not used maybe used used

Levenshtein Distance with Dynamic Programming:

Are all cells in the matrix equally important?

8

i \ j 1 2 3 4 5 1 1 2 2 3 3 4 4 5 5

What is the worst possible distance for two strings of size 5? 5 mismatches. This means all paths of length >5 can be excluded

Levenshtein Distance with Dynamic Programming:

Are all cells in the matrix equally important?

9

i \ j 1 2 3 4 5 1 1 2 2 3 3 4 4 5 5

Paths through red cells have all length >5 Only 24 out of 36 can contribute to results.

not used maybe used used

Are we solving the right problem?

10

Are all alignments useful? Only results with reasonable edit distance. For size 5 strings, let‘s say that‘s 3.

Levenshtein Distance with Dynamic Programming:

Are all cells in the matrix equally important?

11

i \ j 1 2 3 4 5 1 1 2 2 3 3 4 4 5 5

Paths through red cells have all length >3 Only 16 out of 36 can contribute to results.

not used maybe used used

12

Ukkonen E. (1983) On approximate string matching. In: Karpinski M. (eds) Foundations of Computation Theory. FCT 1983. Lecture Notes in Computer Science, vol 158. Springer

Are we Solving the Right Problem?

13

Are we Solving the Right Problem?

• Aligning completely unrelated sequences not needed
• A reasonable alignment has some perfect matches
• BLAST Idea:
• 1. Identify small perfect matches (termed words)
• 2. Extend perfect matches on the same diagonal
• 3. Merge extended perfect matches

(with dynamic programming)

14

BLAST Idea

• Do we need the

alignments for all sequences in the database?

• No, only for “reasonable”

hits introduce a ➞ threshold

• A “reasonable” alignment

will contain short stretches

• f perfect matches
• Find these first, then

extend them to connect them as best possible

15

Perfect Matches (Words) Formally

16

BLAST: Filtration Algorithm

• detects all matching words of length p (of a query

string a in a target string b, both of length n), combining it to an alignment of a and b with no more than k mismatches

• Potential match detection: Find all matches of p-

tuples between a and b (can be done in linear time by inserting them into a hash table)

• Potential match verification: Verify each potential

match by extending it to the left and right until either the first k+1 mismatches are found or the beginning

• r end of the sequences are found

17

BLAST Mechanism Summary

18

word length p (here: p = 4) no mismatches

• r gaps allowed
• nly within the

grey areas

Zipf's Law

19

Example for BLAST

20

SWISS_PROT:C79A_HUMAN P11912 Search SWISSPROT for Ig-alpha:

Example for BLAST

21

Example for BLAST

22

Example for BLAST

23

Distribution of Hits:

Example for BLAST: Alignment

>gi|126779|sp|P11911|C79A_MOUSE B-cell antigen receptor complex associated protein alpha-chain precursor (IG-alpha) (MB-1 membrane glycoprotein) (Surface-IGM-associated protein) (Membrane-bound immunoglobulin associated protein) (CD79A) Length = 220 Score = 278 bits (711), Expect = 5e-75 Identities = 150/226 (66%), Positives = 165/226 (73%), Gaps = 6/226 (2%) Query: 1 MPGGPGVLQALPATIFLLFLLSAVYLGPGCQALWMHKVPASLMVSLGEDAHFQCPHNSSN 60 MPGG + LL LS LGPGCQAL + P SL V+LGE+A C N+ Sbjct: 1 MPGG----LEALRALPLLLFLSYACLGPGCQALRVEGGPPSLTVNLGEEARLTC-ENNGR 55 Query: 61 NANVTWWRVLHGNYTWPPEFLGPGEDPNGTLIIQNVNKSHGGIYVCRVQEGNESYQQSCG 120 N N+TWW L N TWPP LGPG+ G L VNK+ G C+V E N ++SCG Sbjct: 56 NPNITWWFSLQSNITWPPVPLGPGQGTTGQLFFPEVNKNTGACTGCQVIE-NNILKRSCG 114 Query: 121 TYLRVRQPPPRPFLDMGEGTKNRIITAEGIILLFCAVVPGTLLLFRKRWQNEKLGLDAGD 180 TYLRVR P PRPFLDMGEGTKNRIITAEGIILLFCAVVPGTLLLFRKRWQNEK G+D D Sbjct: 115 TYLRVRNPVPRPFLDMGEGTKNRIITAEGIILLFCAVVPGTLLLFRKRWQNEKFGVDMPD 174 Query: 181 EYEDENLYEGLNLDDCSMYEDISRGLQGTYQDVGSLNIGDVQLEKP 226 +YEDENLYEGLNLDDCSMYEDISRGLQGTYQDVG+L+IGD QLEKP Sbjct: 175 DYEDENLYEGLNLDDCSMYEDISRGLQGTYQDVGNLHIGDAQLEKP 220

24

Example for BLAST: Taxonomy

Lineage Report root . cellular organisms . . Eukaryota [eukaryotes] . . . Fungi/Metazoa group [eukaryotes] . . . . Bilateria [animals] . . . . . Coelomata [animals] . . . . . . Gnathostomata [vertebrates] . . . . . . . Tetrapoda [vertebrates] . . . . . . . . Amniota [vertebrates] . . . . . . . . . Eutheria [mammals] . . . . . . . . . . Homo sapiens (man) ---------------------- 473 33 hits [mammals] . . . . . . . . . . Bos taurus (bovine) ..................... 312 2 hits [mammals] . . . . . . . . . . Mus musculus (mouse) .................... 278 31 hits [mammals] . . . . . . . . . . Canis familiaris (dogs) ................. 37 1 hit [mammals] . . . . . . . . . . Rattus norvegicus (brown rat) ........... 35 7 hits [mammals] . . . . . . . . . . Oryctolagus cuniculus (domestic rabbit) . 29 1 hit [mammals] . . . . . . . . . Coturnix japonica ------------------------- 33 2 hits [birds] . . . . . . . . . Gallus gallus (chickens) .................. 31 4 hits [birds] . . . . . . . . Xenopus laevis (clawed frog) ---------------- 30 2 hits [amphibians] . . . . . . . Heterodontus francisci ------------------------ 28 1 hit [sharks and rays] . . . . . . Drosophila melanogaster ------------------------- 30 2 hits [flies] . . . . . Caenorhabditis elegans ---------------------------- 29 1 hit [nematodes] . . . . Saccharomyces cerevisiae (brewer's yeast) ----------- 33 1 hit [ascomycetes] . . . Marchantia polymorpha --------------------------------- 29 1 hit [liverworts] . . Agrobacterium tumefaciens str. C58 ---------------------- 28 1 hit [a-proteobacteria] . Human adenovirus type 3 ----------------------------------- 30 1 hit [viruses] . Human adenovirus type 7 ................................... 30 1 hit [viruses] 25

PSI-Blast

• Globin family (oxygen transport) of proteins occurs in

many species

• Proteins have same function and structure and
• But there are pairs of members of the family sharing

less than 10% identical residues

• A

B C

• PSI-BLAST idea: score via intermediaries may be

better than score from direct comparison

26

50% Only 10% 50%

PSI-BLAST

• PSI-BLAST
• 1. BLAST
• 2. Collect top hits
• 3. Build multiple sequence alignment from significant

local matches

• 4. Build profile
• 5. Re-probe database with profile
• 6. Go back to 2.

27

PSI-BLAST

• But beware of PSI-BLAST:
• False positives propagate and spread through

iterations

• If protein A consists of domains D and E, and protein B
• f domains E and F and protein C of domain F, then

PSI-BLAST will relate A and C although they do not share any domain

28