Ab initio protein modelling for automated X-ray crystal structure solution

蛋白质构成了所有生物的功能机械。它们的特殊作用取决于它们的三维结构，这种三维结构允许给定的蛋白质与其环境中的其他分子进行特定的相互作用。有些蛋白质--酶--走得更远，可以将某些化合物转化为其他化合物。为了更好地了解蛋白质是如何工作的，并能够将其用于工业和医学，科学家们对弄清楚它们的三维结构非常感兴趣。有多种方法可以做到这一点，但占主导地位的技术是X射线结晶学。在这个过程中，一束强烈的X射线被发射到蛋白质晶体上。当X射线穿过晶体时会发生衍射，产生一种具有所研究蛋白质特征的射线图案。为了阐明蛋白质的结构，必须将从同一蛋白质在不同条件下获得的多个衍射图获得的信息汇总在一起。获取这种额外的衍射图可能是耗时、昂贵的，并且通常涉及危险化学品。然而，存在一种技术，计算机通过从类似于正在研究的蛋白质的可用结构中估计等效信息来取代额外的实验。通过这种方式，可以从一个单一的衍射图中解算出蛋白质的结构。这种技术被称为分子置换(MR)，它快速、经济、清洁，而且往往不复杂。然而，由于MR依赖于预先存在的结构，所以它不适用于许多感兴趣的蛋白质，因为对于这些蛋白质来说，类似的结构是根本无法获得的。多年来，科学家们一直试图开发纯粹基于蛋白质序列的计算机方法来预测蛋白质的结构。这些方法通常被称为从头算建模方法。在过去的十年里，这些努力开始取得成果。这些预测模型不太可能在短期内取代晶体结构，因为它们通常包含误差，但最近的研究表明，它们有时足够接近真实结构，可以用于磁共振过程。这就是这项提议背后的主要思想--使目前的从头算建模程序适应先生的具体需要，并使用从头算建模，通常情况下，计算机计算越详细(即越长)，你可以做出的模型就越好。不幸的是，实现最好的模型要求非常高，以至于通常需要大量的计算时间，或者访问超级计算机或其他大量的计算机资源。很少有结晶学家能够使用这些设备，这使得建模方法不切实际。因此，我们提出了一种不同的方法，有效地利用了可以在典型计算机上轻松获得的更简单的模型。在我们的初步工作中，我们已经证明了这种方法可以成功地用于先生，我们现在要做的是找到产生最优模型的最佳方法，并自动完成。这实际上意味着调整该方法以满足现代X射线结晶学的要求，使其变得快速，以便它可以用作常规方法，并被其他结晶学家使用，而不需要从头算建模的专业知识。然后，我们希望将该方法包括在Mr BUMP程序中，这是一个成熟的程序包，允许轻松、自动地将Mr BUMP添加到名为CCP4i的软件包中，该软件包广泛用于结晶学家。通过将我们的处理方法整合到一个熟悉的程序中，我们预计它将在世界各地得到广泛使用。我们预计，通过扩展MR计算方法，我们将能够更快、更便宜地确定蛋白质结构。这样，在药物设计等依赖蛋白质结构信息的各种领域的研究将更快地进行。

Proteins make up the functional machinery of all living beings. Their particular roles depend on their 3-dimensional structures which allow given proteins to interact specifically with other molecules in their environment. Some proteins - enzymes - go further and can transform certain compounds into others. To understand better how proteins work and be able to use them in industry and medicine, scientists are greatly interested in figuring out their 3-dimensional structures. There are various ways to do this, but the dominant technique is X-ray crystallography. In this, an intense beam of X-rays is fired at a protein crystal. The X-rays are diffracted when passing through the crystal, producing a pattern of rays that is characteristic of the protein under study. In order to elucidate the structure of the protein, information derived from multiple diffraction patterns obtained from the same protein but under different conditions must be drawn together. The acquisition of such extra diffraction patterns can be time consuming, expensive, and commonly involves hazardous chemicals. A technique exists, however, where computers substitute the additional experiments by estimating equivalent information from available structures of proteins similar to that under study. In this way, protein structures can be solved from one single diffraction pattern. This technique - called Molecular Replacement (MR) - is fast, economical, clean and often uncomplicated. However, since MR relies on pre-existing structures, it is not applicable to many proteins of interest, for which similar structures are simply not available. For many years, scientists have tried to develop computer methods to predict the structure of proteins, purely based on their sequences. These methods are generally called ab initio modelling methods. Over the past decade, these efforts have started to bear fruit. These predicted models are unlikely to substitute for crystal structures any time soon since they typically contain errors, but recent work has shown that they are sometimes close enough to the real structure for them to be used in the MR process. This is the main idea behind this proposal - to adapt current ab initio modelling procedures to the specific needs of MR. With ab initio modelling, it is generally the case that the more detailed (i.e. the longer) the computer calculation, the better the model you can make. Unfortunately, achieving the best models is so demanding that it often requires extensive calculation times or access to supercomputers or other vast computer resources. Few crystallographers have access to these facilities, making the modelling method impractical. We therefore propose a different approach, making efficient use of simpler models that can be easily obtained on typical computers. In our preliminary work, we have already proven that this approach can work successfully for MR. What we want to do now is find the best way to produce optimal models and to do this automatically. This effectively means adapting the method to meet the demands of modern X-ray crystallography, making it fast so that it can be used as a routine approach and accessible to other crystallographers without specialist knowledge of ab initio modelling. We then want to include the method in the MrBUMP program, which is a well-established package allowing for easy, automated MR. MrBUMP can be added to a software package called CCP4i that is widely used by crystallographers. By incorporating our processing method in a familiar program, we expect it to become widely used across the world. We expect that by extending the MR computational approach we will enable protein structures to be determined more quickly and cheaply. In this way, research in all sorts of areas that depend on protein structure information, like drug design, will proceed faster.