Chen Fu, Sun Huiyong, Liu Hui, Li Dan, Li Youyong, Hou Tingjun
College of Pharmaceutical Sciences, Zhejiang University, Hangzhou, Zhejiang 310058, China.
Phys Chem Chem Phys. 2017 Apr 12;19(15):10163-10176. doi: 10.1039/c6cp08232g.
High-throughput screening (HTS) is widely applied in many fields ranging from drug discovery to clinical diagnostics and toxicity assessment. Firefly luciferase is commonly used as a reporter to monitor the effect of chemical compounds on the activity of a specific target or pathway in HTS. However, the false positive rate of luciferase-based HTS is relatively high because many artifacts or promiscuous compounds that have direct interaction with the luciferase reporter enzyme are usually identified as active compounds (hits). Therefore, it is necessary to develop a rapid screening method to identify these compounds that can inhibit the luciferase activity directly. In this study, a virtual screening (VS) classification model called MIEC-GBDT (MIEC: Molecular Interaction Energy Components; GBDT: Gradient Boosting Decision Tree) was developed to distinguish luciferase inhibitors from non-inhibitors. The MIECs calculated by Molecular Mechanics/Generalized Born Surface Area (MM/GBSA) free energy decomposition were used to energetically characterize the binding pattern of each small molecule at the active site of luciferase, and then the GBDT algorithm was employed to construct the classifiers based on MIECs. The predictions to the test set show that the optimized MIEC-GBDT model outperformed molecular docking and MM/GBSA rescoring. The best MIEC-GBDT model based on the MIECs with the energy terms of ΔG, ΔG, ΔG, and ΔG achieves the prediction accuracies of 87.2% and 90.3% for the inhibitors and non-inhibitors in the test sets, respectively. Moreover, the energetic analysis of the vital residues suggests that the energetic contributions of the vital residues to the binding of inhibitors are quite different from those to the binding of non-inhibitors. These results suggest that the MIEC-GBDT model is reliable and can be used as a powerful tool to identify potential interference compounds in luciferase-based HTS experiments.
高通量筛选（HTS）广泛应用于从药物发现到临床诊断和毒性评估等许多领域。萤火虫荧光素酶通常用作报告基因，以监测高通量筛选中化合物对特定靶点或信号通路活性的影响。然而，基于荧光素酶的高通量筛选的假阳性率相对较高，因为许多与荧光素酶报告酶有直接相互作用的假象物或混杂化合物通常被鉴定为活性化合物（命中物）。因此，有必要开发一种快速筛选方法来识别这些能直接抑制荧光素酶活性的化合物。在本研究中，开发了一种名为MIEC-GBDT（MIEC：分子相互作用能量成分；GBDT：梯度提升决策树）的虚拟筛选（VS）分类模型，以区分荧光素酶抑制剂和非抑制剂。通过分子力学/广义玻恩表面积（MM/GBSA）自由能分解计算得到的MIECs用于从能量角度表征每个小分子在荧光素酶活性位点的结合模式，然后采用GBDT算法基于MIECs构建分类器。对测试集的预测表明，优化后的MIEC-GBDT模型优于分子对接和MM/GBSA重打分。基于具有ΔG、ΔG、ΔG和ΔG能量项的MIECs的最佳MIEC-GBDT模型对测试集中抑制剂和非抑制剂的预测准确率分别达到87.2%和90.3%。此外，对关键残基的能量分析表明，关键残基对抑制剂结合和非抑制剂结合的能量贡献有很大差异。这些结果表明，MIEC-GBDT模型是可靠的，可作为一种强大的工具，用于识别基于荧光素酶的高通量筛选实验中的潜在干扰化合物。
