Department of Biochemistry, University of Oxford, South Parks Road, Oxford, OX1 3QU, UK.
Department of Biostatistics and Medical Informatics, School of Medicine, Acıbadem Mehmet Ali Aydınlar University, İstanbul, 34752, Turkey.
Brief Bioinform. 2020 Dec 1;21(6):2112-2125. doi: 10.1093/bib/bbz143.
MM-PB/GBSA methods represent a higher-level scoring theory than docking. This study reports an extensive testing of different MM-GBSA scoring schemes on two bromodomain (BRD) datasets. The first set is composed of 24 BRPF1 complexes, and the second one is a nonredundant set constructed from the PDBbind and composed of 28 diverse BRD complexes. A variety of MM-GBSA schemes were analyzed to evaluate the performance of four protocols with different numbers of minimization and MD steps, 10 different force fields and three different water models. Results showed that neither additional MD steps nor unfixing the receptor atoms improved scoring or ranking power. On the contrary, our results underscore the advantage of fixing receptor atoms or limiting the number of MD steps not only for a reduction in the computational costs but also for boosting the prediction accuracy. Among Amber force fields tested, ff14SB and its derivatives rather than ff94 or polarized force fields provided the most accurate scoring and ranking results. The TIP3P water model yielded the highest scoring and ranking power compared to the others. Posing power was further evaluated for the BRPF1 set. A slightly better posing power for the protocol which uses both minimization and MD steps with a fixed receptor than the one which uses only minimization with a fully flexible receptor-ligand system was observed. Overall, this study provides insights into the usage of the MM-GBSA methods for screening of BRD inhibitors, substantiating the benefits of shorter protocols and latest force fields and maintaining the crystal waters for accuracy.
MM-PB/GBSA 方法比对接代表了更高层次的评分理论。本研究对两种溴结构域 (BRD) 数据集上的不同 MM-GBSA 评分方案进行了广泛测试。第一组由 24 个 BRPF1 复合物组成，第二组是由 PDBbind 构建的非冗余集，由 28 个不同的 BRD 复合物组成。分析了各种 MM-GBSA 方案，以评估具有不同最小化和 MD 步骤数量的四个方案的性能，使用了 10 种不同的力场和三种不同的水模型。结果表明，增加 MD 步骤或不固定受体原子都不能提高评分或排序能力。相反，我们的结果强调了固定受体原子或限制 MD 步骤数量的优势，不仅可以降低计算成本，还可以提高预测准确性。在所测试的 Amber 力场中，ff14SB 及其衍生物而不是 ff94 或极化力场提供了最准确的评分和排序结果。与其他模型相比，TIP3P 水模型产生了最高的评分和排序能力。进一步对 BRPF1 数据集进行了定位能力评估。观察到使用固定受体的最小化和 MD 步骤的协议比仅使用完全灵活的受体-配体系统的最小化协议具有稍好的定位能力。总体而言，本研究为使用 MM-GBSA 方法筛选 BRD 抑制剂提供了深入了解，证实了较短协议和最新力场以及保持晶体水以提高准确性的优势。
