Using distance geometry algorithms to derive molecular conformational ensembles

动机：当配体（如药物分子）与蛋白质的结合位点对接时，两个分子都会发生构象变化。这些变化很难预测，这使得最终组合结构的必要几何建模变得复杂。为了简化分析，药物设计者可能会做出过于乐观的假设，认为结合位点是刚性的，只考虑配体的灵活性。更复杂的分析涉及到配体和结合位点的灵活性：我们认为结合位点的构象变化是与蛋白质最低能量构象相对应的结构的小偏差。这种“诱导拟合假说”为对接事件提供了更为现实的解释。然而，这并不完全足够，有许多蛋白质-配体复合物的结合位点发生了广泛的构象变化。最近，不同的研究人员设计了更复杂的几何描述，以更全面地捕捉绑定事件的复杂性。这些研究假设蛋白质通常处于低能量构象，但由于热搅拌或拥挤，它可能偶尔采用不同的构象，对应于“折叠漏斗”底部附近的局部能量最小值。这个构象的能量只比全局最小值的能量大一点。这种“构象集合”的物理存在使任何为药物设计工作构建预测机器学习算法的尝试都变得非常复杂，因为很难收集为每种未知构象提供足够样本的训练数据。我们的目标是研究结合位点的几何建模，目的是设计生成构象集合的算法，作为配体对接的前奏。

Motivation: When a ligand (such as a drug molecule) docks with the binding site of a protein, both molecules undergo conformational changes. These changes are difficult to predict and this complicates the necessary geometric modeling of the final combined structure. To simplify the analysis, a drug designer may make the over-optimistic assumption that the binding site is rigid and flexibility is only considered for the ligand. A more complicated analysis deals with the flexibility of both the ligand and the binding site: We consider conformational changes in the binding site to be small deviations from the structure that corresponds to the lowest energy conformation of the protein. This "induced fit hypothesis" provides a more realistic account of the docking event. Nonetheless, it is not fully adequate and there are many protein-ligand complexes in which the binding site has undergone extensive conformational changes. Recently, various researchers have devised more complicated geometrical descriptions in an effort to more fully capture the complexity of a binding event. These studies assume that the protein is usually in a low energy conformation but, due to thermal agitation or crowding, it may occasionally adopt a different conformation that corresponds to a nearby local energy minimum at the bottom of the "folding funnel". This conformation would have an energy that is only slightly larger when compared with the energy of the global minimum. The physical presence of this "conformational ensemble" greatly complicates any attempt to build predictive machine learning algorithms for drug design efforts because it is difficult to collect training data that provides enough samples for each of the unknown conformations. Our goal is to study the geometrical modeling of binding sites with the goal of devising algorithms for the generation of conformational ensembles as a prelude to ligand docking.