Institute of Automatic Control, Silesian University of Technology , 16 Akademicka Street, 44-100 Gliwice, Poland.
Institute of Chemistry, University of Silesia , 9 Szkolna Street, 40-006 Katowice, Poland.
J Chem Inf Model. 2015 Oct 26;55(10):2168-77. doi: 10.1021/acs.jcim.5b00295. Epub 2015 Oct 9.
In a search for new anti-HIV-1 chemotypes, we developed a multistep ligand-based virtual screening (VS) protocol combining machine learning (ML) methods with the privileged structures (PS) concept. In its learning step, the VS protocol was based on HIV integrase (IN) inhibitors fetched from the ChEMBL database. The performances of various ML methods and PS weighting scheme were evaluated and applied as VS filtering criteria. Finally, a database of 1.5 million commercially available compounds was virtually screened using a multistep ligand-based cascade, and 13 selected unique structures were tested by measuring the inhibition of HIV replication in infected cells. This approach resulted in the discovery of two novel chemotypes with moderate antiretroviral activity, that, together with their topological diversity, make them good candidates as lead structures for future optimization.
在寻找新的抗 HIV-1 化学型的过程中，我们开发了一种多步骤基于配体的虚拟筛选 (VS) 方案，将机器学习 (ML) 方法与特权结构 (PS) 概念相结合。在其学习步骤中，VS 方案基于从 ChEMBL 数据库中获取的 HIV 整合酶 (IN) 抑制剂。评估了各种 ML 方法和 PS 加权方案的性能，并将其用作 VS 过滤标准。最后，使用多步骤基于配体的级联对 150 万种商业上可获得的化合物进行了虚拟筛选，并且通过测量感染细胞中 HIV 复制的抑制作用测试了 13 种选定的独特结构。这种方法导致发现了两种具有中等抗逆转录病毒活性的新型化学型，它们的拓扑多样性使它们成为未来优化的良好候选物。
