Accelerating rational drug design with multi-threaded array processors

医药出口每年为英国经济贡献超过120亿英镑（2004年英国贸易和投资数据）。然而，这种成功受到每年成功开发的新药数量下降的威胁（美国食品和药物管理局在1997年批准了230种药物，去年只有28种）。这种急剧下降表明需要新的方法来开发药物。计算化学有可能提供必要的阶跃变化。目前用于计算药物分子与蛋白质宿主的结合亲和力的方法是有效的，但缺乏可靠性，因为它们未能捕获基本的物理学。量子力学提供了相互作用的物理精确表示，但这些计算目前对于商业用途来说太昂贵了。我们将利用我们现有的专业知识来实现在数千个处理器内核上并行的量子力学方法。这将首先通过ClearSpeed plc的100核处理器架构的并行化实现，其次通过包含这种加速技术的节点集群实现。我们将开发的技术将适用于英特尔等公司的紧急多核设计，并有可能为合理的药物设计做出重大贡献。

Pharmaceutical exports contribute over 12 billion to the UK economy each year (UK Trade and Investment figures for 2004). However, this success is threatened by the decline in the number of successful new drugs developed each year (The American Food and Drug Agency approved 230 drugs in 1997 and only 28 last year). This sharp decline indicates the need for new ways to develop pharmaceuticals. Computational chemistry has the potential to deliver the necessary step-change. Currently used methods for computing binding affinities of drug molecules to protein hosts are efficient but lack reliability because they fail to capture the essential physics. Quantum mechanics provides a physically accurate representation of the interactions, but these calculations are currently too computationally expensive for commercial use. We will use our existing expertise to implement quantum mechanical methods parallelized over thousands of processor cores. This will be achieved firstly by parallelization over the 100-core processor architecture of ClearSpeed plc, and secondly over clusters of nodes containing this acceleration technology. The techniques that we will develop will be applicable to emergent multicore designs from companies such as Intel, and have the potential to make a major contribution to rational drug design.