IMAGE-ESTIMATION METHODS FOR AUTOMATED DNA SEQUENCING

这项工作的主要目标是提高DNA的效率， 通过进一步开发和应用自动化数据进行测序， 凝胶阅读器，我们最近开发的估计方法， 限制性片段大小测定的背景。 我们认为，这种方法 显示出扩展序列凝胶可读范围的希望 从而显著提高测序研究的产率 使用高分辨率电泳。 我们的一般做法， 分析电泳凝胶的数据是为了验证和利用 其他方面开发的模型，并酌情开发和 根据需要评估新模型，以描述 生成观测数据并将这些模型应用于估计 设计用于确定最有可能 产生了这些数据。 实现拟议目标的具体目标 工作是：1）开发和/或导入，完善和验证带模型 作为时间和局部序列的函数的迁移率、幅度和形状， 以及基线随空间和时间的变化; 2）开发和评估 碱基调用算法采用来自开发的模型的先验 在aim 1中，并基于两种情况下数据的后验可能性， 荧光发射的高斯和泊松模型; 3）随后 探索随机与符号联合方法的应用 推理，以纳入更复杂的模型，如将 需要分析异常，如压缩; 4）开发和 应用多级（例如符号、轮廓和扫描强度），注释 用于量化算法性能的评估数据库， 关于错误类型及其频率;以及5）尽早启动 算法的传播（"beta测试"）和评估 数据库;随后的全面传播将由 生物医学部计划继续支持我们的实验室 美国国立卫生研究院NCRR研究技术计划。 因为人类的体型 基因组，全序列的确定将不仅取决于发展 一系列低成本、可靠的技术，可以自动化，也可以在 提高这些技术的信息产量的战略。

The broad goal of the proposed work is to advance the efficiency of DNA sequencing by further developing and applying to the data from automatic electrophoretic-gel readers, estimation methods recently developed by us in the context of restriction-fragment sizing. We believe that such methods show promise for extending the readable range of sequence gels significantly, thereby improving the yield from sequencing studies employing high-resolution electrophoresis. Our general approach to the analysis of data from electrophoretic gels is to verify and capitalize on the models developed by others, where appropriate, and to develop and evluate new models as needed to describe the processes operative in generation of the observed data and apply those models to estimation methods designed to determine the phenomenon of interest most likely to have given rise to the data. Specific aims for accomplishing the proposed work are: 1) To develop and/or import, refine, and verify models of band mobility, amplitude, and shape as a function of time and local sequence, and of baseline variation with space and time; 2) To develop and evaluate base-calling algorithms employing priors derived from the models developed in aim 1 and based on the posterior likelihood of the data under both Gaussian and Poisson models for fluorescent emission; 3) To subsequently explore the application of methods for joint stochastic and symbolic inferences in order to incorporate more complex models such as would be needed to analyze aberrancies such as compressions; 4) To develop and apply a multilevel (e.g. symbolic, profile, and scan-intensity), annotated evaluation database for quantifying performance of the algorithms with respect to error types and their frequencies; and 5) To initiate early dissemination ("beta testing") of the algorithms and the evaluation database; subsequent full-scale dissemination would be under the auspices of planned continuation support of our laboratory from the Biomedical Research Technology Program of NCRR, NIH. Because of the size of the human genome, full sequence determination will depend not only on the development of low-cost-reliable technologies that can be automated but also on strategies to improve the information yield from those technologies.