Computational Protein Design Using AND/OR Branch-and-Bound Search - - PowerPoint PPT Presentation

▶
computational protein design using and or branch and
SMART_READER_LITE
LIVE PREVIEW

Computational Protein Design Using AND/OR Branch-and-Bound Search - - PowerPoint PPT Presentation

May 25, 2023 108 likes •671 views

Computational Protein Design Using AND/OR Branch-and-Bound Search Yichao Zhou Yuexing Zhou Jianyang Zeng Institute for Interdisciplinary Information Sciences Tsinghua University Apr, 2015 Yichao Zhou, Yuexing Zhou, Jianyang Zeng

slide-1
SLIDE 1

Computational Protein Design Using AND/OR Branch-and-Bound Search

Yichao Zhou Yuexing Zhou Jianyang Zeng

Institute for Interdisciplinary Information Sciences Tsinghua University

Apr, 2015

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-2
SLIDE 2

What is Structure-Based Protein Design?

Protein structure template Design Algorithm Rotamer Library Energy Functions Backbone Template Toolbox

Ile Pro His · · · Gly Gly Pro Glu Val Gly Ser Asp Pro Ala Ile Trp · · · Ile Ser Ile

1D amino acid sequence

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-3
SLIDE 3

Why We Need Protein Design?

Applications

Figure: New Drug Discovery Figure: Enzyme Optimization Figure: Drug Resistance Prediction Figure: New Biosensor Design

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-4
SLIDE 4

Protein Design as an Optimization Problem

backbone rotamer ir rotamer js E

1

( i

r

) E2(ir, js)

NP Hard!

Energy Function (Optimization Target) ET =

  • ir

E1(ir) +

  • ir
  • js,i<j

E2(ir, js) Self energy of rotamer ir Pairwise energy between rotamer ir and js

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-5
SLIDE 5

Protein Design as an Optimization Problem

backbone rotamer ir rotamer js E

1

( i

r

) E2(ir, js)

NP Hard!

Energy Function (Optimization Target) ET =

  • ir

E1(ir) +

  • ir
  • js,i<j

E2(ir, js) Self energy of rotamer ir Pairwise energy between rotamer ir and js

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-6
SLIDE 6

Search Algorithm

Space Search Algorithm

Exact Algorithm

DEE/A* [Leach98] Branch and Bound [Hong09] Integer Lin- ear Program [Kings- ford05] Tree De- composition [Xu06] …

Approx Algorithm

… Monte Carlo [Voigt00] Simulated Annealing Belief Propagation [Yanover02] Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-7
SLIDE 7

Search Algorithm

Space Search Algorithm

Exact Algorithm

DEE/A* [Leach98] Branch and Bound [Hong09] Integer Lin- ear Program [Kings- ford05] Tree De- composition [Xu06] …

Approx Algorithm

… Monte Carlo [Voigt00] Simulated Annealing Belief Propagation [Yanover02] Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-8
SLIDE 8

Traditional Branch-and-bound Search Space

  • 32
  • 37
  • 37
  • 32
  • 30
  • 30
  • 25
  • 36
  • 35
  • 34
  • 29
  • 31
  • 22
  • 17

⇒ Residue 1 ⇒ Residue 2 ⇒ Residue 3 Heuristic function Heuristic function h(x) returns a lower bound of the energy of leaf nodes in the subtree rooted at x.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-9
SLIDE 9

Traditional Branch-and-bound Search

  • 32
  • 37
  • 37
  • 32
  • 30
  • 30
  • 25
  • 36
  • 35
  • 34
  • 29
  • 31
  • 22
  • 17

Current Best: ∞

⇒ Residue 1 ⇒ Residue 2 ⇒ Residue 3 Heuristic function Heuristic function h(x) returns a lower bound of the energy of leaf nodes in the subtree root at x.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-10
SLIDE 10

Traditional Branch-and-bound Search

  • 32
  • 37
  • 37
  • 32
  • 30
  • 30
  • 25
  • 36
  • 35
  • 34
  • 29
  • 31
  • 22
  • 17

Current Best: ∞

⇒ Residue 1 ⇒ Residue 2 ⇒ Residue 3 Heuristic function Heuristic function h(x) returns a lower bound of the energy of leaf nodes in the subtree root at x.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-11
SLIDE 11

Traditional Branch-and-bound Search

  • 32
  • 37
  • 37
  • 32
  • 30
  • 30
  • 25
  • 36
  • 35
  • 34
  • 29
  • 31
  • 22
  • 17

Current Best: ∞

⇒ Residue 1 ⇒ Residue 2 ⇒ Residue 3 Heuristic function Heuristic function h(x) returns a lower bound of the energy of leaf nodes in the subtree root at x.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-12
SLIDE 12

Traditional Branch-and-bound Search

  • 32
  • 37
  • 37
  • 32
  • 30
  • 30
  • 25
  • 36
  • 35
  • 34
  • 29
  • 31
  • 22
  • 17

Current Best: −37

⇒ Residue 1 ⇒ Residue 2 ⇒ Residue 3 Heuristic function Heuristic function h(x) returns a lower bound of the energy of leaf nodes in the subtree root at x.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-13
SLIDE 13

Traditional Branch-and-bound Search

  • 32
  • 37
  • 37
  • 32
  • 30
  • 30
  • 25
  • 36
  • 35
  • 34
  • 29
  • 31
  • 22
  • 17

Current Best: −37

⇒ Residue 1 ⇒ Residue 2 ⇒ Residue 3 Heuristic function Heuristic function h(x) returns a lower bound of the energy of leaf nodes in the subtree root at x.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-14
SLIDE 14

Traditional Branch-and-bound Search

  • 32
  • 37
  • 37
  • 32
  • 30
  • 30
  • 25
  • 36
  • 35
  • 34
  • 29
  • 31
  • 22
  • 17

Current Best: −37

⇒ Residue 1 ⇒ Residue 2 ⇒ Residue 3 Heuristic function Heuristic function h(x) returns a lower bound of the energy of leaf nodes in the subtree root at x.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-15
SLIDE 15

Traditional Branch-and-bound Search

  • 32
  • 37
  • 37
  • 32
  • 30
  • 30
  • 25
  • 36
  • 35
  • 34
  • 29
  • 31
  • 22
  • 17

Current Best: −37

⇒ Residue 1 ⇒ Residue 2 ⇒ Residue 3 Heuristic function Heuristic function h(x) returns a lower bound of the energy of leaf nodes in the subtree root at x.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-16
SLIDE 16

Traditional Branch-and-bound Search

  • 32
  • 37
  • 37
  • 32
  • 30
  • 30
  • 25
  • 36
  • 35
  • 34
  • 29
  • 31
  • 22
  • 17

Current Best: −37

⇒ Residue 1 ⇒ Residue 2 ⇒ Residue 3 Heuristic function Heuristic function h(x) returns a lower bound of the energy of leaf nodes in the subtree root at x.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-17
SLIDE 17

Traditional Branch-and-bound Search

  • 32
  • 37
  • 37
  • 32
  • 30
  • 30
  • 25
  • 36
  • 35
  • 34
  • 29
  • 31
  • 22
  • 17

Current Best: −37

⇒ Residue 1 ⇒ Residue 2 ⇒ Residue 3 Heuristic function Heuristic function h(x) returns a lower bound of the energy of leaf nodes in the subtree root at x.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-18
SLIDE 18

Traditional Branch-and-bound Search

  • 32
  • 37
  • 37
  • 32
  • 30
  • 30
  • 25
  • 36
  • 35
  • 34
  • 29
  • 31
  • 22
  • 17

Current Best: −37

⇒ Residue 1 ⇒ Residue 2 ⇒ Residue 3 Heuristic function Heuristic function h(x) returns a lower bound of the energy of leaf nodes in the subtree root at x.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-19
SLIDE 19

Traditional Branch-and-bound Search

  • 32
  • 37
  • 37
  • 32
  • 30
  • 30
  • 25
  • 36
  • 35
  • 34
  • 29
  • 31
  • 22
  • 17

Current Best: −37

⇒ Residue 1 ⇒ Residue 2 ⇒ Residue 3 Heuristic function Heuristic function h(x) returns a lower bound of the energy of leaf nodes in the subtree root at x.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-20
SLIDE 20

Traditional Branch-and-bound Search

  • 32
  • 37
  • 37
  • 32
  • 30
  • 30
  • 25
  • 36
  • 35
  • 34
  • 29
  • 31
  • 22
  • 17

Current Best: −37

⇒ Residue 1 ⇒ Residue 2 ⇒ Residue 3 Heuristic function Heuristic function h(x) returns a lower bound of the energy of leaf nodes in the subtree root at x.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-21
SLIDE 21

Traditional Branch-and-bound Search

  • 32
  • 37
  • 37
  • 32
  • 30
  • 30
  • 25
  • 36
  • 35
  • 34
  • 29
  • 31
  • 22
  • 17

Current Best: −37

⇒ Residue 1 ⇒ Residue 2 ⇒ Residue 3 Heuristic function Heuristic function h(x) returns a lower bound of the energy of leaf nodes in the subtree root at x.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-22
SLIDE 22

Traditional Branch-and-bound Search

  • 32
  • 37
  • 37
  • 32
  • 30
  • 30
  • 25
  • 36
  • 35
  • 34
  • 29
  • 31
  • 22
  • 17

Current Best: −37

⇒ Residue 1 ⇒ Residue 2 ⇒ Residue 3 Heuristic function Heuristic function h(x) returns a lower bound of the energy of leaf nodes in the subtree root at x.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-23
SLIDE 23

Traditional Branch-and-bound Search

  • 32
  • 37
  • 37
  • 32
  • 30
  • 30
  • 25
  • 36
  • 35
  • 34
  • 29
  • 31
  • 22
  • 17

Current Best: −37

⇒ Residue 1 ⇒ Residue 2 ⇒ Residue 3 Heuristic function Heuristic function h(x) returns a lower bound of the energy of leaf nodes in the subtree root at x.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-24
SLIDE 24

Residue Interaction Network

Approximation Simplify the energy model by only considering the interaction between residues whose energy term is greater than a threshold. Approx.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-25
SLIDE 25

Problem of Traditional Branch-and-bound

Figure: Residue Interaction Network

+ VS

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-26
SLIDE 26

Problem of Traditional Branch-and-bound

Figure: Residue Interaction Network

+ VS

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-27
SLIDE 27

Problem of Traditional Branch-and-bound

Figure: Residue Interaction Network

+ VS

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-28
SLIDE 28

Problem of Traditional Branch-and-bound

Figure: Residue Interaction Network

+ VS

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-29
SLIDE 29

Problem of Traditional Branch-and-bound

Figure: Residue Interaction Network

+ VS

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-30
SLIDE 30

AND/OR Search Space

A B C (a) Residue Interaction Network

A B 1 C 1 1 B 1 C 1

(b) AND/OR Search Space

Node value vx =   

  • y∈child(x) vy,

if x is an AND node; miny∈child(x) e(y) + vy, if x is an OR node. energy terms between y and ancestors of y

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-31
SLIDE 31

AND/OR Search Space

A B C (a) Residue Interaction Network

A B 1 C 1 1 B 1 C 1

(b) AND/OR Search Space

Node value vx =   

  • y∈child(x) vy,

if x is an AND node; miny∈child(x) e(y) + vy, if x is an OR node. energy terms between y and ancestors of y

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-32
SLIDE 32

AND/OR Search Space

A B C (a) Residue Interaction Network

A B 1 C 1 1 B 1 C 1

(b) AND/OR Search Space

Node value vx =   

  • y∈child(x) vy,

if x is an AND node; miny∈child(x) e(y) + vy, if x is an OR node. energy terms between y and ancestors of y

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-33
SLIDE 33

Finding Sub-optimal Solutions

Problems

AND/OR branch-and-bound search can find GMEC efficiently Protein design models are not perfect:

errors in energy functions and structure of protein templates assumptions of rigid backbone and discrete side-chain conformations

The GMEC solution is not sufficient in practice Sub-optimal (e.g., second-best, …) conformations are required Finding the k best solutions add a threshold to prevent from pruning sub-optimal solutions maintain k node values for each node, i.e., array vx[1], vx[2], . . . , vx[k]

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-34
SLIDE 34

Finding Sub-optimal Solutions

Problems

AND/OR branch-and-bound search can find GMEC efficiently Protein design models are not perfect:

errors in energy functions and structure of protein templates assumptions of rigid backbone and discrete side-chain conformations

The GMEC solution is not sufficient in practice Sub-optimal (e.g., second-best, …) conformations are required Finding the k best solutions add a threshold to prevent from pruning sub-optimal solutions maintain k node values for each node, i.e., array vx[1], vx[2], . . . , vx[k]

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-35
SLIDE 35

Finding Sub-optimal Solutions

Problems

AND/OR branch-and-bound search can find GMEC efficiently Protein design models are not perfect:

errors in energy functions and structure of protein templates assumptions of rigid backbone and discrete side-chain conformations

The GMEC solution is not sufficient in practice Sub-optimal (e.g., second-best, …) conformations are required Finding the k best solutions add a threshold to prevent from pruning sub-optimal solutions maintain k node values for each node, i.e., array vx[1], vx[2], . . . , vx[k]

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-36
SLIDE 36

Finding Sub-optimal Solutions

Problems

AND/OR branch-and-bound search can find GMEC efficiently Protein design models are not perfect:

errors in energy functions and structure of protein templates assumptions of rigid backbone and discrete side-chain conformations

The GMEC solution is not sufficient in practice Sub-optimal (e.g., second-best, …) conformations are required Finding the k best solutions add a threshold to prevent from pruning sub-optimal solutions maintain k node values for each node, i.e., array vx[1], vx[2], . . . , vx[k]

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-37
SLIDE 37

Finding Sub-optimal Solutions

Problems

AND/OR branch-and-bound search can find GMEC efficiently Protein design models are not perfect:

errors in energy functions and structure of protein templates assumptions of rigid backbone and discrete side-chain conformations

The GMEC solution is not sufficient in practice Sub-optimal (e.g., second-best, …) conformations are required Finding the k best solutions add a threshold to prevent from pruning sub-optimal solutions maintain k node values for each node, i.e., array vx[1], vx[2], . . . , vx[k]

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-38
SLIDE 38

Computing v[·] values for OR nodes

x a b c

?

vx[2]

?

vx[1]

?

vx[3]

3

vb[2]

2

vb[1]

6

vb[3]

2

va[2]

1

va[1]

3

va[3]

5

vc[2]

3

vc[1]

10

vc[3] 1 2 2 3 3 3 5 6 10 Problem:

Find k best rotamer assignments for x; Sort all the v values for the child nodes of x; Keep k smallest node values

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-39
SLIDE 39

Computing v[·] values for OR nodes

x a b c

?

vx[2]

?

vx[1]

?

vx[3]

3

vb[2]

2

vb[1]

6

vb[3]

2

va[2]

1

va[1]

3

va[3]

5

vc[2]

3

vc[1]

10

vc[3] 1 2 2 3 3 3 5 6 10 Problem:

Find k best rotamer assignments for x; Sort all the v values for the child nodes of x; Keep k smallest node values

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-40
SLIDE 40

Computing v[·] values for OR nodes

x a b c 2 vx[2] 1 vx[1] 2 vx[3]

3

vb[2]

2

vb[1]

6

vb[3]

2

va[2]

1

va[1]

3

va[3]

5

vc[2]

3

vc[1]

10

vc[3] 1 2 2 3 3 3 5 6 10 Problem:

Find k best rotamer assignments for x; Sort all the v values for the child nodes of x; Keep k smallest node values

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-41
SLIDE 41

Computing v[·] values for AND nodes

x a b c

?

vx[2]

?

vx[1]

?

vx[3]

3

vb[2]

1

vb[1]

6

vb[3]

2

va[2]

1

va[1]

3

va[3]

5

vc[2]

1

vc[1]

10

vc[3] Problem:

Want to find k best ways to merge the sub-problems; vx[i] should be the sum of va[ia] + vb[ib] + vc[ic]; Obviously vx[1] = va[1] + vb[1] + vc[1], hard for vx[2] and vx[3]; Brute force algorithm: O(kd) Our algorithm: O(dk log(dk))

Develop a new method to reduce the complexity exponentially Use binary heap as the data-structure for efficiency

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-42
SLIDE 42

Computing v[·] values for AND nodes

x a b c

?

vx[2]

?

vx[1]

?

vx[3]

3

vb[2]

1

vb[1]

6

vb[3]

2

va[2]

1

va[1]

3

va[3]

5

vc[2]

1

vc[1]

10

vc[3] Problem:

Want to find k best ways to merge the sub-problems; vx[i] should be the sum of va[ia] + vb[ib] + vc[ic]; Obviously vx[1] = va[1] + vb[1] + vc[1], hard for vx[2] and vx[3]; Brute force algorithm: O(kd) Our algorithm: O(dk log(dk))

Develop a new method to reduce the complexity exponentially Use binary heap as the data-structure for efficiency

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-43
SLIDE 43

Computing v[·] values for AND nodes

x a b c

?

vx[2] 3 vx[1]

?

vx[3]

3

vb[2] 1 vb[1]

6

vb[3]

2

va[2] 1 va[1]

3

va[3]

5

vc[2] 1 vc[1]

10

vc[3] Problem:

Want to find k best ways to merge the sub-problems; vx[i] should be the sum of va[ia] + vb[ib] + vc[ic]; Obviously vx[1] = va[1] + vb[1] + vc[1], hard for vx[2] and vx[3]; Brute force algorithm: O(kd) Our algorithm: O(dk log(dk))

Develop a new method to reduce the complexity exponentially Use binary heap as the data-structure for efficiency

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-44
SLIDE 44

Computing v[·] values for AND nodes

x a b c ? vx[2]

3

vx[1] ? vx[3]

3

vb[2]

1

vb[1]

6

vb[3]

2

va[2]

1

va[1]

3

va[3]

5

vc[2]

1

vc[1]

10

vc[3] Problem:

Want to find k best ways to merge the sub-problems; vx[i] should be the sum of va[ia] + vb[ib] + vc[ic]; Obviously vx[1] = va[1] + vb[1] + vc[1], hard for vx[2] and vx[3]; Brute force algorithm: O(kd) Our algorithm: O(dk log(dk))

Develop a new method to reduce the complexity exponentially Use binary heap as the data-structure for efficiency

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-45
SLIDE 45

Computing v[·] values for AND nodes

x a b c

?

vx[2]

3

vx[1]

?

vx[3]

3

vb[2]

1

vb[1]

6

vb[3]

2

va[2]

1

va[1]

3

va[3]

5

vc[2]

1

vc[1]

10

vc[3] Problem:

Want to find k best ways to merge the sub-problems; vx[i] should be the sum of va[ia] + vb[ib] + vc[ic]; Obviously vx[1] = va[1] + vb[1] + vc[1], hard for vx[2] and vx[3]; Brute force algorithm: O(kd) Our algorithm: O(dk log(dk))

Develop a new method to reduce the complexity exponentially Use binary heap as the data-structure for efficiency

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-46
SLIDE 46

Environment

Experiment OSPREY platform from Donald Lab (Duke University) daoopt AOBB framework from UC Irvine Native sequence recovery (core redesign) Large rotamer library for betuer accuracy DEE/A* as the comparison Environment Information CPU: Intel Xeon™ E5-1620 3.6GHz Memory: 4G Memory Limitation

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-47
SLIDE 47

Environment

Experiment OSPREY platform from Donald Lab (Duke University) daoopt AOBB framework from UC Irvine Native sequence recovery (core redesign) Large rotamer library for betuer accuracy DEE/A* as the comparison Environment Information CPU: Intel Xeon™ E5-1620 3.6GHz Memory: 4G Memory Limitation

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-48
SLIDE 48

Experiment

Summary

Dataset 23 previously unsolvable protein redesign cases 5 previously solvable protein redesign cases 6 full protein redesign cases Results solve 21 out of 23 problems that was previously unsolvable; provide a large speedup by several orders of magnitude for previously solvable problems; solve all full protein redesign problems while DEE/A* exceeds 4G memory limitation.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-49
SLIDE 49

Experiment

Summary

Dataset 23 previously unsolvable protein redesign cases 5 previously solvable protein redesign cases 6 full protein redesign cases Results solve 21 out of 23 problems that was previously unsolvable; provide a large speedup by several orders of magnitude for previously solvable problems; solve all full protein redesign problems while DEE/A* exceeds 4G memory limitation.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-50
SLIDE 50

Experiment

Summary

Dataset 23 previously unsolvable protein redesign cases 5 previously solvable protein redesign cases 6 full protein redesign cases Results solve 21 out of 23 problems that was previously unsolvable; provide a large speedup by several orders of magnitude for previously solvable problems; solve all full protein redesign problems while DEE/A* exceeds 4G memory limitation.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-51
SLIDE 51

Experiment

Running Time

Running time of AND/OR branch-and-bound algorithm Running time of our optimized OSPREY Running time of traditional OSPREY Conformation Space

PDB Space size OSPREY cOSPREY AOBB 1IQZ 7.11e+17 1,824,235 40,217 117 2COV 1.14e+10 317 21 1 3FGV 6.44e+12 59,589 5,091 < 1 3DNJ 5.11e+12 7,469 570 3 2FHZ 1.83e+18 3,475,716 70,783 13 Time is measured in millisecond.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-52
SLIDE 52

Conclusion and Future Work

Our Contribution Apply AND/OR branch-and-bound search to protein design Design an algorithm to find sub-optimal solutions Test AOBB search in native sequence recovery experiment Results Memory efficient Speedup by several orders of magnitude comparing to A*/DEE Future Work Using AND/OR best-first search for sub-optimal solutions Parallelizing AOBB on a super computer

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-53
SLIDE 53

Acknowledgement

Discussion

  • Prof. Alexander Ihler
  • Dr. Lars Otuen

Funding National Basic Research Program of China National Natural Science Foundation of China China’s Youth 1000-Talent Program.

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound

slide-54
SLIDE 54

Qvestion and Answer

Thank you!

Yichao Zhou, Yuexing Zhou, Jianyang Zeng Computational Protein Design Using AND/OR Branch-and-Bound