Computing with DNA

James A. Foster Laboratory for Applied Logic Dept. of Computer Science University of Idaho May 8, 1997

Outline part one

Chemistry of DNA Polymerase Chain Reaction Brute Force Computing Finding Hamiltonian Paths

May 8, 1997 jaf

Outline part two

A Mathematical Model Solving SAT P-DNA is PSPACE Potential and Limitations

May 8, 1997 jaf

Chemistry of DNA
DNA molecules: paired strands of nucleotides bases attached to sugarphosphate backbones Nucleotides bases: Adenine binds with Thymine, Guanine binds with Cytosine Backbone: 5 carbons, similar to linked list One molecule's 50 binds to next ones 30 10 binds to nucleotide Paired Strands: Bases bond to complementary strand Sequences: listed 50 to 30
Laboratory for Uphill Computing
May 8, 1997 jaf

DNA Molecule Illustration

5 3

T A

G C

C G

A T

GCCA

TGCATTC

CGGT

ACGTAAG

Note:

GCCATGCATTC

CGGTACGTAAG

s is complement of s is
May 8, 1997 jaf

Laboratory for Uphill Computing

Polymerase Chain Reaction PCR

Given: collection of DNA and two primers, s; t Action: amplify strands of the form
any sequence v
Input: Repeat until satisfied 1 denature DNA with heat 2 anneal DNA with primers 3 elongate strands with DNA polymerase

svt

for

tube T of DNA, primers s and t

Note: copy number for target doubles each iteration

Laboratory for Uphill Computing

May 8, 1997 jaf

0) Given

5 3

PCR Illustration
s s t

1111111111111111 0000000000000000 11 00 1111111111111111 0000000000000000 1111111111111111 0000000000000000 11 00 1111111111111111 0000000000000000

t s t

3 5

1) Denature (heat)

1111111111111111 0000000000000000 11 00 1111111111111111 0000000000000000 11111 00000 111111111111 000000000000 11111 00000 111111111111 000000000000 11111 00000 1111111111 0000000000
t s s t

2) Anneal (Add primers)

s

111111111111111111 000000000000000000 111111111111111111 000000000000000000 11111 00000 111111111111 000000000000 11111 00000 1111111111 0000000000 11111 00000
t t s

11111 00000

3) Elongate (add polymerase)

s s

111111111111111111 000000000000000000 111111111111111111 000000000000000000 11111 00000 111111111111 000000000000 11111 00000 1111111111 0000000000 11111 00000
t t t s

11111 00000
Laboratory for Uphill Computing

May 8, 1997 jaf

Brute Force Computing

To solve problem P :
Represent input instance x as DNA Make tube T Amplify positive results in T Sample T to get answer Represent possible solutions to P x



as DNA

with every possible solution to P x



Recall: NP problems are easy to solve given a short hint". This algorithm checks all possible hints" in parallel, with a polynomial number of operations.

Laboratory for Uphill Computing

May 8, 1997 jaf

Example: Finding Hamiltonian Paths

The NP complete directed Hamitonian Path dHP problem:

Given: Directed graph G, nodes f , t Question: Is there a directed path, visiting every node exactly once, from f to t in G?
2 5

3
A possible Hamitonian path: (F,2,4,6,3,5,T)

Laboratory for Uphill Computing

May 8, 1997 jaf

DNA Algorithm for dHP

Input: Graph G, nodes f and t 0 Represent nodes, edges, paths with DNA 1 Fill tubes with all possible paths 2 Select paths from f to t 3 Select paths of correct length 4 Select paths without duplicate vertexes 5 If anything remains Then return ``yes'' Else ``no''

Laboratory for Uphill Computing

May 8, 1997 jaf

0: Representation Node v: 20 random base pair sequence S

v

Long enough not to bind to each other Short enough for PCR to work Edge u; v: build sequence S with 10-base su x from S , 10-base pre x from S except use all of S and S 
uv u v f t

Path a

b
c:

catenate a; b and b; c

Laboratory for Uphill Computing

May 8, 1997 jaf

10

Example Examples of Adlemans encoding Nodes S2: GTCACACTTC GGACTGACCT S2: AGGTCAGTCC GAAGTGTGAC S4: TGTGCTATGG GAACTCAGCG S4: CGCTGAGTTC CCATAGCACA S5: CACGTAAGAC GGAGGAAAAA S5: TTTTTCCTCC GTCTTACGTG Edges (2,4): GGACTGACCT TGTGCTATGG (4,5): GAACTCAGCG CACGTAAGAC Paths (2.4.5):

GTCACACTTC GGACTGACCT TGTGCTATGG GAACTCAGCG CACGTAAGAC GGAGGAAAAA

11
AGGTCAGTCC GAAGTGTGAC

CGCTGAGTTC CCATAGCACA

TTTTTCCTCC GTCTTACGTG

1: Fill tube with all paths

Amplify tubes of Sx and Sx for each node x Amplify tubes of Suv and Suv for each edge Mix all tubes into tube T

uv

Overlapping segments will bind and leave sticky ends" to promote further binding
Example
Edge (2,4) Node 4
GGACTGACCT TGTGCTATGG CGCTGAGTTC CCATAGCACA

With high probability, every possible path through G will be represented in T

Laboratory for Uphill Computing
May 8, 1997 jaf

12

2,3,4: Select candidate paths f

F T

t

Run PCR on tube T using S and S as primers, put products in T 0 Separate strands with 20n + 10 bases from T 0, put products in tube R Note: by construction, nodes can be visited at most once with high probability If any DNA is left in R, return yes", else no"

Laboratory for Uphill Computing

May 8, 1997 jaf

13

A Mathematical Model
Primitives: tubes of DNA or similar Operations:

Remove T; T 0;
of form
i

,

placing them in T 0

ig

: Remove all strings in T

Detect T : Decide if T has DNA in it Mix  T

f
ig

;T

: Pour all T s into T

i

Copy T;

Tig

: Pour T into each T

We implicitly assume operations such as separation by size, ampli cation, ligation, annealing, and denaturing where needed
Laboratory for Uphill Computing
May 8, 1997 jaf

14

Solving dHP
Representation: as before
Input: for each node v and edge

Tv contains Sv and Sv  0 Tuv contains Suv and Suv  0 MixfTi; Tuv g,T  RemoveT ,T0,fSf g RemoveT0,T 0,fSt g Move length 20n + 10 strings from T 0 to T 00 if DetectT 00
then return ``Yes'' else return ``No''

u; v,

Complexity: linear in number of nodes for Mix Note: S and S  are DNA strands representing node v and edge u; v in input graph
v uv

Laboratory for Uphill Computing

May 8, 1997 jaf

15

Listing Permutations
Problem: input n, list all permutations of items Representation: p1i1p2i : : : p i where codes position j ", i 1; 2; : : : ; n
x n n j 2 f

pj

en-

Input: T with all valid strings for j to n CopyT ,fT1; T2; : : : ; Tng

=1

i is any string other than i MixfT1; T2; : : : ; Tn g,T  first j is are distinct in each string T contains permutations of f1; 2; : : : ; ng
:

for i to n for k j to n RemoveTi,J ,fpj :i; pk ig

=1 = +1

Requires On2 operations

Laboratory for Uphill Computing
May 8, 1997 jaf

16

Solving SAT
The Problem

Given: Boolean formula in CNF p conjunctions, q literals per clause, n variables

^_ l  F ~  = x
p q n i

where l = x or x for some variable x

i;j k k

=1 =1
j

i;j

Question: Is there an ~ s. t. F x~  = T ? x

n n

Example: F ~ 3 = x1 x3  x1 x2 x3  x2 x G~ 2  = x1  x1 x2 x2  x

_ ^ _ _ ^ ^ _ ^

x3

is satis able F T; T; T  = T , G is not

May 8, 1997 jaf

Laboratory for Uphill Computing

17

Laboratory for Uphill Computing

May 8, 1997 jaf

Representing Truth Assignments x1 xn

v1

v2

Etc.

vn

x1

xn

Variables: x1, x2, ..., xn Extra nodes: v1, v2, ..., vn Paths: sequences of literals over these variables

18

DNA Algorithm for SAT

T0 full of all truth assignment p q Input: Boolean formula F = ^i=1_j =1li;j  for i = 1 to p for j = 1 to p if li;j is positive then RemoveTi,1,T 0,fxj g else RemoveTi,1,T 0,fxj g Strands in T 0 will make clause j true Re-label T 0 as Ti if DetectTp
Input: then return ``yes'' else return ``no''

Requires On2 operations

Laboratory for Uphill Computing
May 8, 1997 jaf

19

Computational Complexity
Let P-DNA be problems solvable with polynomial steps in this model Thm Beaver: P-DNA = PSPACE Generalized Turing-complete models splicing systems, DNA TMs exist But, P-DNA computations still require exponential volume, and perhaps lots of clock time

Laboratory for Uphill Computing

May 8, 1997 jaf

20

Disadvantages
Steps" are manual and slow Reaction time proportional to volume of reactants: real time can be much slower than number of steps Required volume can be huge Processes can introduce errors Processes do not scale up well

Laboratory for Uphill Computing

May 8, 1997 jaf

21

Possible solutions to problems

Add active transport or catalyst to tubes Build targeted solutions forget brute force Compute on surfaces Change molecules

Laboratory for Uphill Computing

May 8, 1997 jaf

22

Advantages
Massive parallelism Attack special instances e.g. Keller graphs for MC Very low energy consumption 10,19 J versus 10,9 J per basic operation with no inherent lower bound Way cool

Laboratory for Uphill Computing

May 8, 1997 jaf

23

L. Adleman, Molecular Computation of Solutions to Combinatorial Problems", Science, 266:1021 1024, 1994 L. Adleman, On Constructing a Molecular Computer", manuscript, ftp: [Link] pub csinfo papers adlemanmolecular [Link], 1995 D. Beaver, A Universal Molecular Computer", Technical report, Penn. State U., 1995 J. Hartmanis, On the Weight of Computations", Bull. Euro. Assoc. for Theoretical Comp. Sci., 55:136 138, 1995 J. H. Reif, Parallel Molecular Computation", in Proc. 7th ACM Symp. on Parallel Alg. and Arch., pp. 213 223, 1995 Also see URLs from my homepage bookmarks

Further Reading

Laboratory for Uphill Computing

May 8, 1997 jaf

24