Inhibox.com

Cloud Computing: A Drug
Discovery Game Changer?
Cloud computing has the potential to transform drug discovery in many areas – heralding the dawn of no-compromise computer-aided drugdiscovery and making effective virtual screening a reality.
Automation versus craftsmanship? High-throughput
teething problems, of course. How did you know the screening was supposed to be the answer to declining
compound being tested was reliably what it claimed to be pharmaceutical research productivity, but many feel it
on the tin? How could you handle that data deluge to failed to deliver the benefits it initially promised. This is
tease out useful knowledge? But the market responded, so often the case with a new technology. These great ideas
as markets do, and solutions appeared. HTS took its come along, shake our world and then we wait for the
place as a $2 billion market in mainstream drug promised revolution. Sometimes it comes through – often
discovery, but still can be said ultimately to have failed to it does not. My question is: will cloud computing disrupt
deliver on its early promise. Let’s look at some of the the drug discovery game, or just speed it up a little?
reported statistics that underpin that proposition: My belief is that its effect will be transformational – ◆ The estimated costs associated with HTS in an here’s why. Let’s start with a refresher, and make the link average pharmaceutical drug discovery project are between high-throughput screening, computer-aided around $1 million. Pharmaceutical companies each drug discovery and cloud computing along the way. spend, on average, $50 million per year on HTS (1) ◆ A majority of HTS campaigns fail to find a lead MODELLING & SCREENING
Back in the 1980s, the advent of computer-aided drug discovery (CADD) promised to revolutionisepharmaceutical research. Teams of expert modellers were brought together in dedicated groups in all of the world’smajor pharmaceutical, agrochemical and biotech BRINGING IT ALL TOGETHER
companies. Software suppliers vied for leadership in the – VIRTUAL SCREENING
space, and hardware suppliers made a killing from selling The world of CADD was quick to recognise the value of powerful graphics workstations and compute servers. In the principle of HTS and soon a computer-driven these heady days, rational drug design was proclaimed by equivalent appeared – virtual screening. The principle some as heralding a new dawn in the industry. There was simple: obtain or build a 3D model of the target site were success stories (the anti-flu drug Relenza, the cyclic and dock into it as many candidate structures as you can urea anti-AIDS treatments, and so on) to encourage get hold of to estimate how well they might bind; select things along. The craftsmen modellers remained in their the most promising candidates and buy or make these for teams, serving the organisation, but the vision of all physical screening. The cost of modelling the whole medicinal chemists relying on CADD technology never process is negligible compared with physical HTS, so it came to pass. Twenty-five years on, CADD still must be a valuable pre-cursor and the return on constitutes an enormously valuable component in drug investment should be large – right? Wrong – if you discovery research – but it has not torn up the rule book.
consider the effectiveness of the first wave of virtualscreening operations. Then in the 1990s, along came high-throughputscreening (HTS). Automation had arrived and would The virtual screening principle is sound, though, so we completely revolutionise drug discovery, overturning should examine its failings to date. There are three rational design by either medicinal chemists or modellers. It was a brilliant proposition: “Let’s make/buyand screen everything we can get our hands on and see what sticks!” Robots were bought, compound inventories amassed, and off we charged. There were 3. Simplistic, compromised modelling tools There are three accessible sources of compound data for want at any given time; and they are fully managed by the virtual screening: supplier catalogues, (around 6 million provider. The user needs nothing but a personal computer entries), each company’s additional internal inventory (less and internet access. Significant innovations in than one million and lacking diversity) and publicly virtualisation and distributed computing, as well as available data sets (not necessarily subject to rigorous quality improved access to high-speed internet and a weak control). Considering the universe of possible candidates for economy, have accelerated interest in cloud computing.
modelling, this is a very meagre set of data sources.
The flexibility of the cloud computing resource, allocatedon demand, is a compelling alternative to the huge Virtual screening should not be a totally automated internal resources that would be needed to cope with the process, at least not with the software tools currently peak requirement typical of computational chemistry.
available off the shelf. Not enough of the knowledge ofskilled practitioners has yet been encoded in the The business model for CADD required the purchase of commercially available software for the job. The range of the most powerful compute servers and graphics systems targets and challenges facing the modellers is wide, and you could afford, along with licences for a wide range of their skill and experience is of paramount importance in expensive modelling software platforms, writing the cost determining the outcome of the virtual screening study. off over a given period – usually three years. This wasextremely capital intensive, so the investments were Which brings us to the modelling tools themselves. The constrained. This asset was utilised as needed by the platforms in use today were developed over the last three modellers, which meant it was sometimes churning jobs decades and built on algorithms that optimised through at maximum capacity – while the users queued performance for the systems available at the time. Let’s be patiently to get their work started – and at other times it clear about what this means. Moore’s Law has accurately sat idling, waiting for the next job. At any given foreseen the doubling of computer processing power on a moment, the modellers could have used 10-times its less than two-year cycle for the last four decades. A system power, or did not need it at all. This lumpy usage is designed just 10 years ago (new, in CADD terms) would have been built to deliver the most accurate results it couldwithin a ‘reasonable’ time, on hardware with three per cent Cloud computing acts to lift the business constraint so of the power of today’s equivalents. Scientists in industry that there is, in effect, an immediate jump in the power can be impatient souls, and rightly so; thus ‘reasonable’ available, over and above the evolutionary improvements means up to a day for a large task, and a minute for a small driven by Moore’s Law. So now all of those nasty software activity along the way. Go beyond these limits, and design compromises should come into question. Cloud compromises have to be made in the algorithm or its computing ought to herald the dawn of no-compromise parameters to make the time taken acceptable. Modelling computer-aided drug discovery and, specifically, it can is a very complex, multivariate job – so this meant lots of make effective virtual screening a reality. Virtual deeply-seated compromises, embedded in the systems and screening fits the cloud computing model perfectly the way they are applied. March forward ten years and because it is so inherently parallel: you pay no extra for what would have taken a month could be achieved in a using 1,000 CPUs for one hour, rather than waiting 100 day with the same software – so you don’t have to hours for 10 CPUs to do the same job.
One only has to look at some examples to understand the potentially dramatic effect of cloud computing in pharma. CLOUD COMPUTING
Cloud computing offers companies virtually unlimited
Take, for example, the treatment of the receptor site in computing resource on tap – you effectively rent as much virtual screening. The site is flexible and may well be time and power as you need at any given occasion.
occupied with bound water molecules. Add in theflexibility of each candidate drug, and the complexity Cloud computing is a general term for anything that grows alarmingly. Few virtual screening approaches even involves delivering services over the internet; the name begin to attempt to account for such complexities. Yet comes from the symbol often used to represent the issues such as the flexible-ligand and flexible-receptor internet in flowcharts and diagrams. Cloud services are docking have to be considered according to Dr Garrett sold on demand, typically by the hour; they are elastic – Morris, co-author of AutoDock (3), the world’s most a user can have as much or as little of a service as they widely used docking package. AutoDock is used in internet-distributed biomedical grid computing projects Cloud computing is beginning to gain acceptance in such as IBM’s World Community Grid project, pharmaceutical and life sciences companies, with GSK, FightAIDS@Home (4). He and Paul Finn, CSO at Pfizer, Eli Lilly & Co, Johnson & Johnson and Genentech InhibOx, are pioneering cloud-enabled approaches using all quoted as using cloud computing resources.
an internal version of AutoDock and proprietarysoftware to bring new, necessary rigour to the art. They IS CLOUD COMPUTING A GAME CHANGER?
have built a 100 million compound database of flexible I think it is. And it’s not a question of automation versus 3D candidate models, with calculated shape, craftsmanship – it’s a combination of the two. stereochemistry (5), charge and physical properties toguide the virtual screening (6). It is massively computer- The scientists behind developing CADD systems have intensive, but it is appropriately rigorous and cloud always been grasping for more computing power, to make practical their developments. Ten years ago,Professor Graham Richards at Oxford University was The visionaries have seen the potential for cloud leading the screen-saver project (9), pulling spare CPU computing to transform drug discovery in many areas – power from over three million PCs across the world to beyond the creation of effective virtual screening. drive virtual screening of cancer drug candidates. Cloudcomputing makes that amazing vision a practical, Dave Powers, Lilly’s Associate Information Consultant commercial reality. At last, computing power is available for Discovery IT, was recently quoted as describing a and cost-effective enough to cause the design of project the firm executed using Amazon’s Elastic modelling systems to be fundamentally reassessed. It Compute Cloud (EC2) service as follows (7): “We were should herald the development of no-compromise recently able to launch a 64-machine cluster computer CADD and effective virtual screening – and that could working on bioinformatics sequence information, indeed change the drug discovery game.
complete the work and shut it down in 20 minutes. Itcost $6.40. To do that internally – to go from nothing to References
getting a 64-machine cluster installed and qualified – is a Industrialization of Drug Discovery,
Handen JS (Ed), CRC Press, 2005
Chemogenomics in Drug Discovery, Methods and
There are already many examples of bioinformatics tools Principles in Medicinal Chemistry, Vol 22
being deployed on EC2. These include HMMer for Morris GM, Huey R, Lindstrom W et al, AutoDock4
protein sequence analysis and BLAST for general and AutoDockTools4: Automated docking with
biological sequence analysis; in fact, pre-built Amazon selective receptor flexibility, J Comput Chem
EC2 images are publicly available for these tools.
Dec, 30 (16): pp2,785-2,791, 2009
Researchers at the Biotechnology and Bioengineering 4. FightAIDS@Home
Center at the Medical College of Wisconsin have created http://www.worldcommunitygrid.org/
a scalable virtual proteomics data analysis cluster research/faah/overview.do
(VIPDAC) that exploits cloud computing services, and Armstrong MS, Morris GM, Finn PW et al, Molecular
they now distribute a pre-configured Amazon Machine similarity including chirality, J Mol Graph Model 28:
Image (AMI) containing the OMSSA and X!Tandem pp368-370, 2009
search algorithms and sequence databases (8).
Ballester PJ, Finn PW and Richards WG, Ultrafast
Shape Recognition: Evaluating a new ligand-based
virtual screening technology, J Mol Graph Model
Paul Davie is Chief Executive Officer at InhibOx Ltd, an Oxford
27: pp836-845, 2009
University Chemistry Department spin-out, pioneering theapplication of cloud computing to computer-aided drug Mullin R, The New Computing Pioneers, Chemical
discovery. He has a long and successful track record in & Engineering News 87 (21): pp10-14, 2009
commercial roles in computer-aided drug discovery. He held Halligan BD, Geiger JF, Vellejos AK et al, Low
support and sales roles at Chemical Design before going on to Cost, Scalable Proteomics Data Analysis Using
build and manage the European sales, marketing, support and Amazon’s Cloud Computing Services and Open
consulting operations at Oxford Molecular. Paul went on Accelrys to establishtheir Consulting Division and serve as European General Manager, before Source Search Algorithms, J Proteome Res 8 (6),
becoming Chief Operating Officer at InforSense. He then founded and was CEO pp3,148-3,153, 2009
at Secerno, a successful database security company, before returning to research Richards WG, Virtual screening using GRID
informatics with his consulting company, Davinger. Paul has an MBA and read computing: the Screensaver Project, Nature
Chemistry at Oxford University. Email: paul.davie@inhibox.com Reviews Drug Discovery 1:pp551-555, 2002

Source: http://www.inhibox.com/sites/default/files/CloudComputingADrugDiscoveryGameChanger.pdf