Joint Bayesian analysis of single-molecule colocalization images and kinetics

项目摘要 目前生物学的后基因组时代的一个中心问题是在分子水平上理解 蛋白质和RNA机器执行所有细胞功能的化学和物理机制 做手术吧。多波长单分子荧光共定位技术(“COSMOS”) 单分子光谱学)方法已被广泛采用并用于阐明功能 大范围的大分子机器的机理，从单个电机酶到 核糖体和剪接体。然而，高效和准确的COSMOS数据分析，特别是大型、 多维数据集，仍然具有挑战性。COSMOS数据集本质上很难分析，因为 在需要避免的有限激发强度下对热驱动单分子过程的观测 光漂白本质上是噪声和随机的，因此将受益于基于 优化统计理论，得出准确的结论。 该应用为基于贝叶斯图像的COSMOS数据分析提出了一种新的方法 分类、贝叶斯马尔可夫链蒙特卡罗等基于统计的方法。整个项目 目标是产生比现有方法更准确、易于扩展的分析方法 大型数据集，更可靠，即使在经验较少的用户手中也是如此。特别是，我们将 开发算法和实现软件，以1)充分利用两个文件中包含的信息- 维度宇宙图像，2)使用客观的、统计上严格的方法来计算 每个图像中存在一个给定的分子物种，3)将动力学分析与图像分类相结合 允许基于现有数据得出关于分子机制的最准确的结论，以及4)消除 人工分析和主观参数调整在现有分析方法中引入了偏差。这个 开发的模型和算法将通过针对广泛范围的全面测试进行改进和验证 模拟的和已知结果的经验数据集。该项目的具体目标是：1)加强 与时间无关的贝叶斯斑点识别算法及其性能表征 时变联合贝叶斯判别/隐马尔可夫模型的实现与表征 (BD/HMM)算法从COSMOS数据中推导出分子机制，以及3)开发和分发 用于贝叶斯宇宙图像分析的可用的、有文档记录的开源软件包。

Project Summary A central concern of the present post-genomic era of biology is understanding at the molecular level the chemical and physical mechanisms by which the protein and RNA machines that perform all cellular functions operate. Multi-wavelength single-molecule fluorescence co-localization techniques (“CoSMoS”; co-localization single-molecule spectroscopy) methods have been widely adopted and used to elucidate the functional mechanisms of a broad range of macromolecular machines ranging from individual motor enzymes to the ribosome and spliceosome. However, efficient and accurate CoSMoS data analysis, particularly of large, multi-dimensional datasets, remains challenging. CoSMoS datasets are inherently difficult to analyze because observations of thermally-driven single-molecule processes at the limited excitation intensities needed to avoid photobleaching are intrinsically noisy and stochastic and thus would benefit from objective methods based on optimized statistical theory to derive accurate conclusions. This application proposes a new approach to CoSMoS data analysis based on Bayesian image classification, Bayesian Markov chain Monte Carlo, and other statistics-based methods. The overall project goal is to produce analytical methods that are more accurate than existing approaches, readily scalable to large datasets, and are more reliable, even in the hands of less experienced users. In particular, we will develop algorithms and implement software that will 1) make full use of the information contained in the two- dimensional CoSMoS images, 2) use objective, statistically rigorous approaches to calculate the probability of a given molecular species being present in each image, 3) integrate kinetic analysis with image classification to allow the most accurate conclusions about molecular mechanisms based on available data, and 4) eliminate the manual analysis and subjective parameter tweaking that introduce bias in existing analytical methods. The developed models and algorithms will be refined and validated through thorough testing against a broad range of simulated and known-outcome empirical data sets. The specific aims of the project are to: 1) enhance the Time-Independent Bayesian Spot Discrimination algorithm and characterize its performance, 2) develop, implement and characterize a time-dependent Joint Bayesian Discrimination/Hidden Markov Modeling (BD/HMM) algorithm to derive molecular mechanisms from CoSMoS data, and 3) develop and distribute a usable, documented, open-source software package for Bayesian CoSMoS image analysis.