Accelerating molecular docking on Graphics Processors

Simon McIntosh-Smith Richard B. Sessions University of Bristol, UK simonm@cs.bris.ac.uk, R.Sessions@bristol.ac.uk

1

!  Drug docking examples: Elastase inhibitors

2

!  Prion disease Prion protein behind Creutzfeld-Jacob disease in humans and shown here binding with a (pink) porphyrin-based ligand The porphyrin's bound iron ion is just showing in yellow

3

1,719 atoms in the protein 53 atoms in the ligand

!  BUDE: Bristol University Docking Engine

Typical docking scoring functions

Empirical Free Energy Forcefield

BUDE

Free Energy calculations MM1,2 QM/MM3

Entropy: solvation configurational Electrostatics All atom Explicit solvent

No Yes Yes Approx Approx Yes ? Approx Yes No Yes Yes No No Yes

Accuracy

Speed

1. MD Tyka, AR Clarke, RB Sessions, J. Phys. Chem. B 110 17212-20 (2006)

2. MD Tyka, RB Sessions, AR Clarke, J. Phys. Chem. B 111 9571-80 (2007)

3. CJ Woods, FR Manby, AJ Mulholland, J. Chem. Phys. 128 014109 (2008) 4

!  Empirical Free Energy Function (atom-atom)

ΔGligand binding = i=1∑

Nprotein

j=1∑Nligand

f(xi,xj)

Parameterised using experimental data†

† N. Gibbs, A.R. Clarke & R.B. Sessions, "Ab-initio Protein Folding using Physicochemical Potentials and a Simplified Off-Lattice Model", Proteins 43:186-202,2001 5

!  Redocking into Xray Structure

6 1CIL (Human carbonic anhydrase II)

!  Another example

7 1EZQ (Human Factor XA)

!  Self docking results (Ligand Binding Database)

8

0 2 4 6 8

10 12 14 16 18 20

3b65

3g

0w

1hfs

3e

92

2fzz

1z

95

1mq6

3i

5z

2hvc

1m

q5

2nw

4 3d

83

3d7z

1k

v2

1ezq

2p

16

2cji

2i0d

3e

93

1hfc

2y

xj

3ibu

1w

82

2qo8

2i

s7

1nfx

1f

js

1nfw

3e

5a

2adu

1g

cz

3ibn

1c

il 3b

yz

2hoc

1i

91

2oaz

1s

2q

2rbe

3b

zu

1mnc

1u

sn

2pvh

2g

v7

1jg0

1l

qd

1tsl

3g

9l

2zb0

3i

6o

2hb3

2b

q6

0

5

10

15

20

3i5z

1k

v2

3d7z

3e

92

3b65

3g

0w

2cji

1z95

1h

fs

2nw

4 2p

16

2fzz

1w

82

1mq6

2h

vc

3d83

2y

xj

3e5a

1m

q5

2i0d

1f

js

1nfx

1h

fc

2oaz

3i

bu

3e93

1n

fw

1ezq

1c

il 2q

o8

2is7

2p

vh

2rbe

2a

du

3ibn

3b

yz

1gcz

2g

v7

3i6o

2h

b3

1s2q

3b

zu

3g9l

1j

g0

2hoc

1i

91

1mnc

1u

sn

1tsl

1l

qd

2bq6

2z

b0

RMSD

Best pose (by binding energy)

Best pose in top 100 poses (by binding energy)

!  What are GPUs and OpenCL all about?

GPU

CPU CPU A modern computer includes: – One or more CPUs – One or more GPUs

9

OpenCL (Open Compute Language) lets programmers write a single portable program that uses ALL

resources in the heterogeneous platform

GPU

!  BUDE Acceleration with OpenCL

GA – like,

energy minimisation

Geometry

(transform ligand)

Energy

i=1∑Nprot

j=1∑Nlig

f(xi,xj)

( ) Rx Rxy Rxz Tx Ryx Ry Ryz Ty Rzx Rzy Rz Tz

Energy of pose

START (input)

(output) END

Copy protein and ligand coordinates

(once)

Host Processor PCI Express Bus GPU accelerator 10 "Benchmarking energy efficiency, power costs and carbon emissions on heterogeneous systems", Simon McIntosh-Smith, Terry Wilson,

Amaurys Avila Ibarra, Jon Crisp and Richard B.Sessions, The Computer Journal, September 12th 2011. DOI: 10.1093/comjnl/bxr091

!  Systems benchmarked High-end: •  Supermicro 1U dual

GPU server •  Two Intel 5500 series

2.4 GHz Xeon ‘Nehalem’ quad-core processors

•  Two Nvidia C2050 ‘Fermi’ GPUs or

•  Two AMD ‘Cypress’ FirePro V7800s

Medium-end: •  Workstation with 1

CPU & 1 GPU •  Intel E8500 3.16 GHz

dual core CPU •  Previous generation

Nvidia consumer-level GPU, the GTX280

11

!  Supermicro GPU server

12

!  Systems benchmarked Middle-end: •  Workstation based on

a 3-core AMD 2.8 GHz Phenom II X3 720

•  4 GBytes of DRAM •  No GPU!

Low-end: •  Laptop based on an

Intel Core2Duo SU9400 ‘Penryn’ 1.4 GHz CPU

•  4 GBytes of DRAM •  No GPU!

13

!  Benchmarking methodology

14

•  Use the same power measurement equipment for all the systems under test

•  Watts Up? Pro meter •  Measures complete system

power‘at the wall’ •  Run as fast as possible on all available

resources (i.e. all cores or all GPUs simultaneously)

!  Relative performance

15

162 seconds per simulation

1,120 seconds per simulation

Less than 2 days to screen a library of 1 million drug candidates on 1000 GPUs

6.9X speedup

!  Relative energy efficiency

16

0.011 kWh per simulation

0.034 kWh per simulation

0.011 kWh = 0.16 pence per simulation 1 million simulations GPUs save £3,680

on energy costs for one experiment

68% energy reduction for the same work

!  Future work •  Modify the code to use the CPUs and

GPUs at the same time (done) •  Further optimisations (done – code

already several times faster) •  Refine forcefield (in progress … forever) •  Improvements to the Genetic Algorithm

and further trimming of the search space (done)

17

!  For an introduction to GPUs The GPU Computing Revolution – a Knowledge Transfer Report from the London Mathematical Society and the KTN for Industrial Mathematics •  https://ktn.innovateuk.org/web/mathsktn/

articles/-/blogs/the-gpu-computing-revolution

18

!  Conclusions •  GPUs can yield big performance increases if your

application has enough of the right sort of parallelism •  GPUs can can also result in significant savings in

energy costs (and equipment costs) •  OpenCL can work just as well for multi-core CPUs as

it does for GPUs

It’s possible to screen libraries of millions of molecules against complex targets using highly accurate, computationally-expensive methods in one weekend using equipment costing << £1M

19

!  References •  S. McIntosh-Smith, T. Wilson, A.A. Ibarra, J.

Crisp and R.B. Sessions, "Benchmarking energy efficiency, power costs and carbon emissions on heterogeneous systems", The Computer Journal, September 12th 2011. DOI: 10.1093/comjnl/bxr091

•  N. Gibbs, A.R. Clarke & R.B. Sessions, "Ab-initio Protein Folding using Physicochemical Potentials and a Simplified Off-Lattice Model", Proteins 43:186-202,200

20