Dynamic Programming (cont’d) CS 466 Saurabh Sinha

RNA secondary structure prediction

RNA RNA is similar to DNA chemically. It is usually only a single strand. T(hyamine) is replaced by U(racil) Some forms of RNA can form secondary structures by “pairing up” with itself. This can change its properties dramatically. linear and 3D view:

RNA There’s more to RNA than mRNA RNA can adopt interesting non-linear structures, and catalyze reactions tRNAs (transfer RNAs) are the “adapters” that implement translation

Secondary structure Several interesting RNAs have a conserved secondary structure (resulting from base- pairing interactions) Sometimes, the sequence itself may not be conserved for the function to be retained It is important to tell what the secondary structure is going to be, for homology detection

Conserved secondary structure N-Y A N-N’ R N-N’ / N Consensus binding site for R17 phage coat protein. N = A/C/G/U, N’ is a complementary base pairing to N, Y is C/U, R is A/G Source: DEKM

Basics of secondary structure G-C pairing: three bonds (strong) A-U pairing: two bonds (weaker) Base pairs are approximately coplanar

Basics of secondary structure

G-C pairing: three bonds (strong) A-U pairing: two bonds (weaker) Base pairs are approximately coplanar Base pairs are stacked onto other base pairs (arranged side by side): “stems”

Secondary structure elements Loop: single stranded subsequences bounded by base pairs loop at the end of a stem stem loop single stranded bases within a stem … only on one side of stem … on both sides of stem

Non-canonical base pairs G-C and A-U are the canonical base pairs G-U is also possible, almost as stable

Nesting Base pairs almost always occur in a nested fashion If positions i and j are paired, and positions i’ and j’ are paired, then these two base-pairings are said to be nested if: i < i’ < j’ < j OR i’ < i < j < j’ Non-nested base pairing: pseudoknot

Pseudoknot (9, 18) (2, 11) NOT NESTED

Pseudoknot problems Pseudoknots are not handled by the algorithms we shall see Pseudoknots do occur in many important RNAs But the total number of pseudoknotted base pairs is typically relatively small

Secondary structure prediction Find the secondary structure with most base pairs. Nussinov’s algorithm Recursive: finds best structure for small subsequences, and works its way outwards to larger subsequences

Nussinov’s algorithm: idea There are only four possible ways of getting the best structure for subsequence (i,j) from the best structures of the smaller subsequences (1) Add unpaired position i onto best structure for subsequence (i+1,j) i i+1 j

Nussinov’s algorithm: idea There are only four possible ways of getting the best structure for subsequence (i,j) from the best structures of the smaller subsequences (2) Add unpaired position j onto best structure for subsequence (i,j-1) j j-1i

Nussinov’s algorithm: idea There are only four possible ways of getting the best structure for subsequence (i,j) from the best structures of the smaller subsequences (3) Add (i,j) pair onto best structure for subsequence (i+1,j-1) j i+1j-1 i

Nussinov’s algorithm: idea There are only four possible ways of getting the best structure for subsequence (i,j) from the best structures of the smaller subsequences (4)Combine two optimal substructures (i,k) and (k+1,j) i kk+1j

Nussinov RNA folding algorithm Given a sequence s of length L with symbols s 1 … s L. Let  (i,j) = 1 if s i and s j are a complementary base pair, and 0 otherwise. We recursively calculate scores g(i,j) which are the maximal number of base pairs that can be formed for subsequence s i …s j. Dynamic programming

Recursion Starting with all subsequences of length 2, to length L g(i,j) = max of g(i+1, j) g(i,j-1) g(i+1,j-1) +  (i,j) max i < k < j [g(i,k) + g(k+1,j)] Initialization g(i,i-1) = 0 g(i,i) = 0 O(n 2 ) ? No. O(n 3 )

Traceback As usual in sequence alignment ? Optimal sequence alignment is a linear path in the dynamic programming table Optimal secondary structure can have “bifurcations” Traceback uses a pushdown stack

Traceback Push (1,L) onto stack Repeat until stack is empty: pop (i,j) if i >= j continue else if g(i+1,j) = g(i,j) push (i+1,j) else if g(i,j-1) = g(i,j) push (i,j-1) else if g(i+1,j-1) +  (i,j) = g(i,j) record (i,j) base pair push (i+1,j-1) else for k = i+1 to j-1, if g(i,k)+g(k+1,j) g(i,j) push (k+1,j) push (i,k) break (for loop)