COMPARING MATHEMATICAL METHODS TO DETECT HIV RESISTANCE

比较检测 HIV 耐药性的数学方法

• 批准号: 6810115
• 负责人: ALEXANDER R MACALALAD
• 金额: $ 8.04万
• 依托单位: HARVARD SCHOOL OF PUBLIC HEALTH
• 依托单位国家: 美国
• 财政年份: 2003
• 资助国家: 美国
• 起止时间: 2003-09-15 至 2005-09-14
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/6810115
• 关键词: RNA directed DNA polymerase; antiAIDS agent; antiviral agents; drug resistance; endopeptidases; gene mutation; genotype; human immunodeficiency virus; informatics; mathematical model; model design /development; molecular biology information system; phenotype; postdoctoral investigator; statistics /biometry; virus classification; virus genetics

DESCRIPTION (provided by applicant): Objective: Mathematical methods such as neural networks and classification trees are being used more frequently in medicine to analyze high dimensional data such as genetic sequences. They have also been applied to the detection of HIV resistance to antiretrovirals by analyzing mutations in the sequences of the target proteins, reverse transcriptase and protease. This study will build accurate classification models to detect HIV resistance, and compare their performance in correctly detecting resistance and in finding meaningful relationships between specific mutation patterns and resistance. Methods: Protease and reverse transcriptase sequences paired with phenotypic resistance data will be taken from the Stanford HIV Sequence Database, a public database of HIV sequences published in the medical literature. Baseline data will be measured, and a systematic review will be performed to describe any variability between studies. Models will be trained to classify sequences as resistant or susceptible to a specific antiretroviral using logistic regression, neural networks, and classification trees. Meta-modeling will be done to explore sources of heterogeneity. The ability of these models to detect resistance from genotype will be compared to each other using measures of bias, discrimination, and calibration. Resistance mutations and mutation patterns found by the models will be compared to what is already known and well accepted in the literature.

描述（由申请人提供）： 目的：神经网络和分类树等数学方法在医学中越来越频繁地用于分析基因序列等高维数据。它们还被应用于通过分析靶蛋白、逆转录酶和蛋白酶序列中的突变来检测HIV对抗逆转录病毒药物的耐药性。这项研究将建立准确的分类模型来检测艾滋病毒耐药性，并比较它们在正确检测耐药性和寻找特定突变模式与耐药性之间有意义的关系方面的表现。 研究方法：与表型耐药数据配对的蛋白酶和逆转录酶序列将取自斯坦福大学HIV序列数据库，该数据库是医学文献中发表的HIV序列的公共数据库。将测量基线数据，并进行系统性综述，以描述研究之间的任何变异性。将使用逻辑回归、神经网络和分类树对模型进行训练，以将序列分类为对特定抗逆转录病毒药物具有抗性或敏感性。将进行元建模以探索异质性的来源。这些模型检测基因型耐药性的能力将使用偏倚、区分度和校准指标进行相互比较。通过模型发现的耐药突变和突变模式将与文献中已知和公认的进行比较。