Statistical mechanics with quantum potentials: Application to protein-ligand binding affinities

在药物开发过程中，先导化合物优化需要密集的化学合成和测试 努力该过程本质上是高度迭代的，需要多轮合成，因为 为改善例如药物动力学因素如溶解度而进行的改变也会降低效力， 需要进一步的变化来恢复效力，等等。 蛋白质-配体结合亲和力将显著降低这种昂贵和耗时的负担， 为药物化学家提供了更积极地优先考虑配体用于合成和基于测试的能力， 计算结果。然而，目前，在蛋白质-配体结合中实现一致的准确性， 亲和性预测是计算化学领域中未实现的目标。常规对接和评分 已经显示了在化学文库中提供活性配体相对于非活性配体的富集的方法，但是仍然 它们通过其结合亲和力对候选配体进行排序的能力非常有限。即使是像自由能源这样的进步 微扰（FEP）和VeraChem自己的挖掘最小自由能方法VM 2，仍然局限于他们的 能够始终如一地提供所需的准确度。重要的是，所有这些方法都有一个共同点， 依赖于经典分子力学（MM）力场，甚至是蛋白质和 药物样分子不能保证具有最佳参数，也不能提供对 涉及例如π堆积、极化、电荷转移或金属阳离子的化学相互作用。事实上， 典型力场中固有的近似被认为是限制精度的关键因素。在这种快速- 跟踪SBIR提案，我们的目标是通过整合VeraChem的自由能方法VM 2来解决这一关键限制 与量子力学（QM）的潜力，产生一个新的软件包QM为基础的蛋白质配体 称为PLQM-VM 2的自由能计算。该软件包将不同于其他自由能方法，如 作为FEP，这是不容易实现的QM潜力。同样，尽管QM已应用于 蛋白质-配体系统，现有的方法仅限于关注单一构象，而PLQM- VM 2将整合现有的基于力场的构象搜索与QM能量和自由能 精致。第一阶段将提供第一水平的QM蛋白质-配体自由能能力，将VM 2与 配体和蛋白质活性位点的快速半经验QM处理。在第二阶段， 并且将通过SEQM/极化力添加超出活性位点的蛋白质原子的精确包含 场方法，和一个非常有效的QM碎片方案将使能量校正在更高级别的QM。 CPU和GPU上的并行化将提供足够快的周转时间，以支持行业研发， 将支持向本地计算机集群和云资源提交计算结果。包装 将通过应用于多个蛋白质靶点进行测试和最佳实践定义，每个靶点都具有高质量 测量了大量非共价抑制剂的亲和力。

During the drug development process, lead optimization requires intensive chemical synthesis and testing efforts. The process can be highly iterative in nature with multiple rounds of synthesis required, because changes made to improve, for example, pharmacokinetic factors such as solubility can also decrease potency, requiring further changes to recover potency, and so on. Consistently accurate computational predictions of protein-ligand binding affinities would significantly reduce this expensive and time consuming burden, by providing medicinal chemists the ability to more aggressively prioritize ligands for synthesis and testing based on computational results. However, currently, achievement of consistent accuracy in protein-ligand binding affinity prediction is an unmet goal in the field of computational chemistry. Conventional docking and scoring methods have been shown to provide enrichment of active vs. inactive ligands in chemical libraries, but still are very limited in their ability to rank candidate ligands by their binding affinities. Even advances like free energy perturbation (FEP) and VeraChem's own mining minima free energy method VM2, remain limited in their ability to consistently provide the accuracy levels needed. Importantly, all of these methods have in common a dependency on classical molecular mechanics (MM) force fields, and even the best force fields for proteins and drug-like molecules are not guaranteed to have optimal parameters nor to provide adequate descriptions of chemical interactions involving, for example, π-stacking, polarization, charge transfer, or metal cations. In fact, the approximations inherent in typical force fields are thought to be a key factor limiting accuracy. In this fast- track SBIR proposal, we aim to address this key limitation by integrating VeraChem's free energy method VM2 with quantum mechanical (QM) potentials, producing a new software package for QM based protein-ligand free energy calculations called PLQM-VM2. This package will be distinct from other free energy methods, such as FEP, which is not readily implemented with QM potentials. Similarly, although QM has been applied to protein-ligand systems, existing methods are limited to focusing on a single conformation, whereas PLQM- VM2 will integrate existing force field-based conformational searching with QM energy and free energy refinement. Phase I will provide a first level of QM protein-ligand free energy capability, integrating VM2 with a fast semi-empirical QM treatment of the ligand and protein active site. In Phase II, a capability to allow fast and accurate inclusion of protein atoms beyond the active site will be added through a SEQM/polarizable force field method, and a very efficient QM fragmentation scheme will enable energy corrections at higher-level QM. Parallelization on CPUs and GPUs will provide fast enough turnaround to support industry R&D, and submission of calculations to both local computer clusters and cloud resources will be supported. The package will be tested and best practices defined through application to multiple protein targets each with high quality measured affinities for a large series of non-covalent inhibitors.