Addressing Open Challenges of Computational Genome Annotation

我们建议利用生物信息学之间正在进行的合作的成功， 格赖夫斯瓦尔德大学（德国）和格鲁吉亚理工学院（美国）的团队 并解决计算基因组注释中的开放挑战。在此过程中 开发，我们计划实施新的算法思想，并满足无偏见的需求， 整合不同类型的OMICS数据。 我们计划解决一个长期存在的问题，在接口的生物信息学和 机器学习-基因发现的自动生成和判别参数化 算法目前结合OMICS证据的方法经常导致预测不足 或者过度预测的工具。充分理解了困难和性质， 不同类型的OMICS证据，我们提出了一种优化的方法，以充分无监督， 生成性和辨别性训练。 我们将引入新的方法来优化多个OMICS证据整合到基因中， 预测.这些想法将进一步发展蛋白质家族为基础的基因发现实施 在AUGUSTUS-PPX。我们建议创建用于基因发现的蛋白质家族的表示 首次包含了跨物种的基因结构信息。 我们将开发一种新的方法，将统一两个先进的研究领域-转录 从RNA-Seq和整合RNA-Seq和同源性的统计基因发现重建 信息.我们将描述一个新的，全面的模型和EM类算法技术 (the“整体”方法），以确定最适合的转录本集及其表达水平 现有的OMICS证据。 我们还将开发一个自动基因发现算法的全部内容的宏基因组 包括真核和病毒宏基因组序列。这项任务通常被认为是 太有挑战性了我们提出了一种利用和推进算法思想的解决方案， 我们在为原核生物宏基因组创造基因发现者的过程中掌握的方法 以及真核生物的基因组。 所有新工具都将在开源许可证下提供给社区。

We propose to capitalize on success of ongoing collaboration between the bioinformatics teams at the University of Greifswald (Germany) and at the Georgia Institute of Technology (USA) and address open challenges in computational genome annotation. In the course of this development, we plan to implement new algorithmic ideas and satisfy the needs of unbiased integration of different types of OMICS data. We plan to address one of the long-standing problems at interface of bioinformatics and machine learning – automatic generative and discriminative parameterization of gene finding algorithms. Current methods of combining OMICS evidence frequently result in under predicting or over predicting tools. Having good understanding of the difficulties and the properties of different types of OMICS evidence we propose an optimized approach to the full unsupervised, generative and discriminative training. We will introduce novel means to optimize integration of multiple OMICS evidence into gene prediction. These ideas will develop further the protein family-based gene finding implemented in AUGUSTUS-PPX. We propose to create representations of protein families for gene finding that for the first time include cross-species gene structure information. We will develop a new approach that will unify two advanced research areas - transcript reconstruction from RNA-Seq and statistical gene finding that integrates RNA-Seq and homology information. We will describe a new, comprehensive model and EM-like algorithmic technique (the “wholistic” approach) to identify the sets of transcripts and their expression levels that best fit the available OMICS evidence. We will also develop an automatic gene-finding algorithm for a full content of metagenomes including eukaryotic and viral metagenomic sequences. This task is conventionally considered too challenging. We propose a solution exploiting and advancing algorithmic ideas and approaches that we mastered in the course of creating gene finders for prokaryotic metagenomes as well as eukaryotic genomes. All new tools will be available to the community under open source licenses.