Bioinformatics Techniques for analysis and comparison of biological

Bioinformatics

Techniques for analysis and comparison of biological data sequences n n n Approximate Pattern Matching Multiple Sequence Alignment Applications in Molecular Biology Problems

Basic definitions (α) n Edit Distance: for 2 strings we define as edit distance the minimum number of operations that are needed in order to transform the first string to the second. The basic operations are insertion, deletion and replacement of symbols. n Example: S 1: vintner and S 2: writers edit-distance(S 1 ->S 2)=5 n

Basic definitions (β) n Edit Transcript: for the transformation of a string, we define the sequence of operations that are needed in order to transform the first string to the second. The basic operations are: Insertion: I ¨ Deletion: D ¨ Replacement: R ¨ Matching: Μ ¨ n n Example: S 1: vintner and S 2: writers edit-distance(S 1 ->S 2)= RIMDMDMMI

Example of Substitution Matrix

Sequence Alignment n Sequence Alignment: we put the one sequence below the other so that common characters are put in the same positions.

Sequence alignment allowing gaps n We align two sequence by introducing 7 gaps in 4 positions, that are translated to changing the DNA sequence in the respective positions.

Computing the score-function when aligning sequences n Score-function= 0+1 -2+0+3+3 -1=4

The method of Dynamic Programming n Dynamic programming: consider 2 sequences S 1 and S 2, we define as D(i, j) the edit distance between prefixes S 1[1. . i] and S 2[1. . j], that is the minimum number of operations that are needed in order to transform the i first characters of S 1 to the j first characters of S 2. n Use 3 basic techniques: ¨ recurrence ¨ tabular relation, computation, ¨ traceback.

Example of Dynamic Programming Table

Recurrence relationship n Recurrence relationship: D(i, j)=min[D(i-1, j)+1, D(i, j-1)+1, D(i-1, j-1)+t(i, j)] ¨ D(i, j-1)+1: we must insert S 2[j] ¨ D(i-1, j)+1: we must delete S 1[i], ¨ D(i-1, j-1)+1: in order to map S 1[i] to S 2[j] we must replace S 1[i], with S 2[j], ¨ D(i-1, j-1): matching

Example: recurrence relationship

Traceback relationship n Traceback: ¨ From (i, j) to (i, j-1) αν D(i, j)= D(i, j-1)+1 (character insertion) ¨ From (i, j) to (i-1, j) if D(i, j)= D(i-1, j)+1 (character deletion) ¨ From (i, j) to (i-1, j-1) if D(i, j)= D(i-1, j-1)+t(i, j) (character replacement or matching)

Adding traceback pointers

Interpreting traceback pointers

Complexity of Dynamic Programming Methods n Initialization: O(n) + O(m) n Recurrence relationship: O(n*m) n Traceback pointers: Ο(n+m) n Complexity: Ο(n 2) n Equivalence with a problem of graph theory where every node has as label (i, j)

Basic definitions (γ) n Weighted Edit Distance: the minimum number of operations that must be done to transform the first string to the second. Every operation has a specific weight-cost. Assume that the basic operations have the following weights: ¨ Insertion or deletion: d ¨ Replacement : r ¨ Matching : m. n n Example : S 1: vintner and S 2: writers weighted edit-distance(S 1 ->S 2)= r+4 d+4 m.

Recurrence relationship with weights n Recurrence relationship: D(i, j)=min[D(i-1, j)+d, D(i, j-1)+d, D(i-1, j-1)+t(i, j)], n where: ¨ t(i, j)= e, if S 1(i)=S 2(j), ¨ t(i, j)=r, if S 1(i) S 2(j) and ¨ D(i, 0)=i*d and D(0, j)=j*d.

Dynamic programming & sequence similarity based on alphabet n Recurrence relationship for sequence similarity : V(i, j)= max[V(i-1, j-1)+s(S 1(i), S 2(j)), V(i-1, j)+ s(S 1(i), _), V(i, j 1)+ s(_, S 2(j))], n where: ¨ s(x, y): the alignment value of character x and y ¨ V(0, j)= s(_, S 2(k)), 1<k<j and V(i, 0)= s(S 1(k), _), 1<k<i.

EXTENTIONS n Weighted Edit Distance. n In Molecular Biology applications the character subsitution weights are stored in substitution matrices - Substitution Matrix: PAM και BLOSUM n It is better to face it as alignment and embed the score. n Every match is positive , everything else is 0 or negative

Extentions n n n Longest Common Subsequence (match weight 1, otherwise 0) End-space free variant that encourages one string to align in the interior of the other, or the suffix of one to align with the prefix of the other (initial conditions 0, everything is countable in the last row and the last column) – shotgun sequence assembly Approximate occurrence of P in T (the optimal alignment of P to a substring of T has distance δ from the optimal alignment) (initial conditions 0). -- locate a cell (n, j) with value greater than δ. -- traverse backpointers from (n. j) to (0, k). -- occurrence in T[k, j]

Local Suffix Alignment Problem n Local suffix alignment problem: for two sequence S 1 and S 2 locate a suffix α of S 1[1. . i] (with the probability of being empty) and a suffix β of S 2[1. . j] (probably empty) so that V(α, β) has the maximum value of all possible pairs of suffices of S 1[1. . i] and S 2[1. . j]. We symbolize as υ(i, j) the maxuimum local suffix alignment for the values i and , j ( i<n and j<m). n Initial conditions v(i, 0)=0 and v(0, j)=0 as we can select an empty suffix. n υ(i, j)=max[0, υ(i-1, j-1)+s(S 1(i), S 2(j)), υ(i-1, j)+ s(S 1(i), _), υ(i, j-1)+ s(_, S 2(j))]

Locally Similar Strings T: S:

Remarks n Global Alignment is called Needleman-Wunsch alignment n Local alignment is called Smith-Waterman alignment n Smith-Waterman can find regions with high similarity by simply performing a trace-back from any cell (i, j) backwards and locate a pair with similarity v(i, j).

Sequence alignment with gaps n Gap: contiguous spaces, we want to control the distribution of gaps. n Introducing gaps: In order to compute the gap introduces in aligning 2 sequences, we can with a simple approach consider that every gap contribute a constant weight Wg, independent of its length. n Value of alignment for “k” gaps: n A better approach is using a function that is a function of the gap length. Then we can fill a matrix: V(i, j)=max[E(i, j), F(i, j), G(i, j)]

Alignment with arbitrary gap weights i j G(i, j)= V(i-1, j-1)+cost(i->j) E(i, j)= max. KV(i, k) - w(j-k) (0<=k<=j-1) i gaps j F(i, j)= maxl{V(l, j) - w(i-l) i (0<=l<=i-1) gaps j

n n n n V(i, j)=max{E(i, j), F(i, j), G(i, j)} V(i, 0)=-w(i) V(0, j)=-w(j) E(i, 0)=-w(i) F(0, j)=-w(j) G(0, 0)=0 Assuming that |S 1|=n and |S 2|=m the recurrences can be evaluated in O(nm 2+n 2 m)

Applications in Molecular Biology problems n Multiple sequence alignment problem: a multiple global sequence of k>2 strings S={ S 1, S 2, …. , Sκ} is a physical generalization of aligning for two strings.

Why we are interested in multiple sequence alignment n Multiple sequence alignment is used: ¨ to identify and represent protein families and super-families, ¨ to represent characteristics that are transferred to DNA sequences or protein families, ¨ To represent the evolution history (phylogenetic trees) in DNA or protein sequences.

Multiple sequence alignment n n Kind of alignments: ¨ Extention of DP approach (too costly) ¨ Use of pairwise alignment (center star algorithm) Algorithms of multiple sequence alignemnt: ¨ FASTA ¨ BLAST.

Sequence Database Searching Βήματα καθορισμού πρωτεϊνικής ακολουθίας 1. Compare the new sequence with PROSITE and BLOCKS in order to locate well-characterized sequence motifs. 2. Search in DNA and protein sequence databases (Genbank, Swiss-Prot, etc. ) for locating sequences that are locally similar (when using a local similarity criterion) – using FASTA and BLAST 3. If these searches provide interesting results, then use the technique of dynamic programming. 4. When we need to involve amino acid substitution matrices, we usually use a variant of the Dayhoff PAM matrix and BLOSUM matrix.

BLAST Algorithm §BLAST: Basic Local Alignment Search Tool, Altschul et. Al. 1990 §Basic idea: locate common sub-sequences of the same length (segment pairs) that appear in the input query sequence and the set of data base sequences. In the sequel extend in order to locate maximal segment pairs Αλγόριθμος Είδος query Είδος sequence BLASTP Πρωτείνη Πρωτεϊνη BLASTN Νουκλεοτίδιο BLASTX Νουκλεοτίδιο Πρωτεϊνη TBLASTN Πρωτεϊνη Νουκλεοτίδιο TBLASTX Νουκλεοτίδιο

O Αλγόριθμος FASTA n n FAST: Fast – All, Lipman et al. 1985 Central idea: use of small words (words ή k-tuples) that appear in both sequences. In the case of protein sequences the word length is το 1 -2 residues while for DNA sequences the word length can reach 6 bases

Algorithm FASTA n n 1 o step: we search for words of length ktup in the dynamic programming table: ‘hot spots’ (pairs (i, j) ) 2ο step: we locate the ten better diagonal runs– diagonal runs από ‘hot-spots’ στον πίνακα (a hot spot define the (i-j)-diagonal, the score is the sum of scores of hot-spots plus weighted decreasing as the distance increases) 3ο step: we align «good sub-alignments» 4ο step: we produce the best path

Άλλα Θέματα q q q Amino Acid Substitution Matrices Dayhoff PAM (Point Accepted Mutation) BLOSUM (derived from the BLOCKS database) BLOCKS is a database of protein motifs that attempts to represent the most highly conserved substrings in aminoacid sequences or related preoteins.

Indexing Approaches n k-gram indexing n direct indexing n vector space indexing

k-gram Indexing Hash tables (FASTP, FASTA, BLAST, BL 2 SEQ, PSI-BLAST, Mega. BLAST, BLASTZ, WU-BLAST, BLAT, SSAHA, SENSEI) q ed-tree (the gram size k, the skip interval Δ, the segment length vector H=[h 1, …, ht], where sumi(hi)=k) q -- For each sequence, the algorithm generates all k-grams with Δ skips -- Each gram is partitioned according to H, and the partitions are inserted into the edtree. The search algorithm partitions accordingly the query string, and initiates a search procedure, beginning from the root. Some important properties: (1) the k-grams are usually longer than that of BLAST, (2) they allow inexact k-gram matches, (3) only one k-gram out of Δ k-grams is indexed, Ed-tree is used by CHAOS, LAGAN, DIALIGN.

Direct Indexing q Suffix trees -- MUMmer, AVID, REPuter, MGA, QUASAR q VP-trees (Vantage Point tree)

MUMmer q Detecting MUM (Maximal Unique Matches), that is pair of subsequences (x’, y’) that exactly match and there is no other matching pair that contains x’, y’ simultaneously (just use the GST(x, y)) q Find the backbone of the alignment: all the pairs (x’, y’) are sorted in increasing order of the position of x’. Next the longest sequence of MUMs whose subsequences from x, y are in sorted order is found. These form the backbone q Closing gaps. The gaps between consecutive MUMs are aligned with the help of Smith-Waterman Similar tools are AVID, REPuter, MGA.

VP-trees (Vantage Point tree) The VP-tree has been adapted to sequence databases where the distance is the edit distance or the block edit distance. It can be applied to other distance functions as long as they are almost metric. The algorithm takes a database, D={s 1, …, sn}, and chooses a sequence s as the root, while the median of the distances to s is computed. The two sets are: (i) the sequences that are closer to s than the median, (ii) the rest of the sequences. Given a query q, the sequences at distance r are found as follows: First, q is compared to the vantage sequence s, at the root node. Let M be the median distance. 1. 2. 3. If d(q, s)≤r, s is inserted to the result set If d(q, s)≤r+M, then the left child is searched recursively. If d(q, s)≥M-r, then the right child is searched recursively.

Vector Space Indexing These index structures, map sequences or subsequences to vectors in a vector space. There exist two important index structures: (i) SST (Sequence Search Tree) (i) MRS (Multi Resolution String) index

SST q. Vector Space Mapping (window size w, shift amount Δ, tuple size k) The parameter k, determines the size of the computed vector (for alphabet size σ this vector has size σκ q. The produced vectors are stored in a so called centroid structure. A query is performed as follows: a query sequence is first divided into subsequences of window size w, using a shift amount of Δ=w/2. Each of the produced query vectors is then searched on the index structure starting from the root node. q