TRANSLATION OF AUTOMATED SEUENCER DATA TO DNA SEQUENCES

这项研究的主要目标是改进自动分析 基于凝胶的DNA测序梯形图，基于模式识别 将原始仪器数据转换为DNA序列。我们强调的是神经 网络，适应特定的测序条件和仪器。这个 模式识别和常规基准化软件的性能 将评估：(1)作为实验错误挑战的描述 人腺病毒基因组的天然等位基因多样性；(2)用于检测 以及二倍体模板实验中杂合基因座的规范； (3)用于大规模测序的引物选择和组装操作 项目。基本定位误差的分布将在 相邻核苷酸同一性的上下文以及不同的结果 排序策略。 模式识别有望带来三大优势 基本计算软件：(1)低聚物痕迹的上下文阵列分析 提高基本呼叫精度；(2)专门负责、神经网络和 算法处理器支持在线信号调理和基本呼叫 实时；(3)信号调理和模式识别 模块支持对每个基本呼叫进行客观的信心衡量。 该项目将显著和积极地影响实现 人类基因组计划的既定目标。无增量成本 需要硬件或战略上的修改。节省的成本可以 通过劳动密集型审查和编辑的自动化实现 主要数据。实时基本呼叫支持更高的吞吐量 仪器，利用更快的分离更大的并行阵列 排序梯子。目标基本呼叫可信度参数支持 序列组装过程中的重叠分配，应便于 通过扩展的数据库进行序列匹配搜索。

The primary objective of this research is to improve automated analysis of gel-based DNA sequencing ladders, through pattern recognition-based translation of raw instrument data to DNA sequences. We emphasize neural networks, adapted to particular sequencing conditions and instruments. The performances of pattern recognition and conventional basecalling software will be evaluated: (1) as experimental errors challenge description of the natural allelic diversity of human adenoviral genomes; (2) for detection and specification of heterozygous loci in diploid template experiments; (3) for primer selection and assembly operations of large scale sequencing projects. Distributions of basecalling errors will be analyzed in the contexts of neighboring nucleotide identities and as results of different sequencing strategies. Three principal advantages are expected from pattern recognition basecalling software: (1) analysis of contextual arrays of oligomer traces improves basecalling accuracy; (2) specifically tasked, neural network and algorithmic processors support on-line signal conditioning and basecalling in real time; and (3) the signal conditioning and pattern recognition modules support objective measures of confidence for each basecall. This project will significantly and positively impact progress towards the stated goals of the human genome initiative. No incremental costs for hardware or strategic modifications are required. Cost savings can be realized through automation of labor intensive review and editing of primary data. Real-time basecalling supports higher throughput instruments, exploiting faster separation of larger parallel arrays of sequencing ladders. Objective basecall confidence parameters support overlap assignment during sequence assembly, and should facilitate sequence - match searches through expanding databases.