Identification of Pathogenic Protein Mutations using Synthetic Biology, Structural Bioinformatics and Biochemistry

该项目的目的是使用一种称为深度突变扫描（DMS）的技术来识别在泛素信号蛋白中产生致病表型的突变。DMS将饱和诱变与深度测序相结合，允许对数千种不同的突变等位基因进行并行的高通量功能分析。重点将放在单个氨基酸的变化，但多个突变对单个蛋白质的潜在上位效应也将被研究。多位点突变体比单位点突变体要多得多，因此DMS实验只覆盖有限的数量。这些数据将被用作机器学习方法的基础，以训练计算预测器来估计剩余上位相互作用的表型效应。结果将与现有的计算表型预测，特别关注那些突变没有确定的DMS experiment.Finally，突变体的子集将进行详细的结构研究进行了研究。本节的目标是尝试和理解为什么表型的计算预测因子通常是不准确的，并进一步了解泛素信号转导的破坏如何导致疾病状态。最终将产生DMS数据集，这可以帮助预测未知影响的泛素信号转导变体。该数据集还将用于构建计算表型预测器，并且可以与其他数据集集成以训练更通用的蛋白质表型预测器。

The aim of this project is to identify mutations which produce a pathogenic phenotype in ubiquitin signalling proteins using a technique called Deep Mutational Scanning (DMS). DMS integrates saturation mutagenesis with deep sequencing, allowing high throughput functional analysis of thousands of different mutant alleles in parallel. The focus will be on single amino-acid changes, but the potential epistatic effects of multiple mutations on a single protein will also be studied.There are exponentially more multi-site mutants than single-site ones, so only a limited number will be covered by the DMS experiment. This data will be used as the basis for a machine-learning approach in order to train a computational predictor to make estimates of the phenotypic effect of the remaining epistatic interactions. The results will be compared to existing computational phenotype predictors, with particular focus on those mutations which were not identified by the DMS experiment.Finally, a subset of mutants will be studied in detail by undertaking structural studies. The goal of this section is to try and understand why computational predictors of phenotypes are often inaccurate, and to gain further insight into how disruption of ubiquitin signalling leads to a disease state.Ultimately a DMS dataset will be produced, which can help make predictions about ubiquitin signalling variants of unknown effect. The dataset will also be used to build a computational phenotype predictor and can be integrated with other datasets to train more general protein phenotype predictors.