Computational Intelligence Group, Departamento de Inteligencia Artificial, Facultad de Informática, Universidad Politécnica de Madrid, Boadilla del Monte 28660, Spain.
Artif Intell Med. 2013 Mar;57(3):219-29. doi: 10.1016/j.artmed.2012.12.005. Epub 2013 Feb 1.
Our aim is to use multi-dimensional Bayesian network classifiers in order to predict the human immunodeficiency virus type 1 (HIV-1) reverse transcriptase and protease inhibitors given an input set of respective resistance mutations that an HIV patient carries.
Multi-dimensional Bayesian network classifiers (MBCs) are probabilistic graphical models especially designed to solve multi-dimensional classification problems, where each input instance in the data set has to be assigned simultaneously to multiple output class variables that are not necessarily binary. In this paper, we introduce a new method, named MB-MBC, for learning MBCs from data by determining the Markov blanket around each class variable using the HITON algorithm. Our method is applied to both reverse transcriptase and protease data sets obtained from the Stanford HIV-1 database.
Regarding the prediction of antiretroviral combination therapies, the experimental study shows promising results in terms of classification accuracy compared with state-of-the-art MBC learning algorithms. For reverse transcriptase inhibitors, we get 71% and 11% in mean and global accuracy, respectively; while for protease inhibitors, we get more than 84% and 31% in mean and global accuracy, respectively. In addition, the analysis of MBC graphical structures lets us gain insight into both known and novel interactions between reverse transcriptase and protease inhibitors and their respective resistance mutations.
MB-MBC algorithm is a valuable tool to analyze the HIV-1 reverse transcriptase and protease inhibitors prediction problem and to discover interactions within and between these two classes of inhibitors.
我们旨在使用多维贝叶斯网络分类器，根据 HIV 患者携带的各自耐药突变，预测人类免疫缺陷病毒 1 型（HIV-1）逆转录酶和蛋白酶抑制剂。
多维贝叶斯网络分类器（MBC）是专门设计用于解决多维分类问题的概率图形模型，其中数据集的每个输入实例都必须同时分配给多个输出类变量，而这些变量不一定是二进制的。在本文中，我们引入了一种新的方法，称为 MB-MBC，通过使用 HITON 算法确定每个类变量周围的马克夫毯，从数据中学习 MBC。我们的方法应用于从斯坦福 HIV-1 数据库获得的逆转录酶和蛋白酶数据集。
就抗逆转录病毒联合疗法的预测而言，与最先进的 MBC 学习算法相比，实验研究在分类准确性方面显示出了有前途的结果。对于逆转录酶抑制剂，我们分别获得了 71%和 11%的平均和全局准确性；而对于蛋白酶抑制剂，我们分别获得了超过 84%和 31%的平均和全局准确性。此外，MBC 图形结构的分析使我们能够深入了解逆转录酶和蛋白酶抑制剂及其各自耐药突变之间的已知和新的相互作用。
MB-MBC 算法是分析 HIV-1 逆转录酶和蛋白酶抑制剂预测问题以及发现这两类抑制剂内部和之间相互作用的有价值的工具。
