機械学習と遺伝的アルゴリズムを用いたエンハンサーの同定とモデル化

使用机器学习和遗传算法的增强剂识别和建模

• 批准号: 17F17797
• 负责人: 中井 謙太
• 金额: $ 0.96万
• 依托单位: The University of Tokyo
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for JSPS Fellows
• 财政年份: 2017
• 资助国家: 日本
• 起止时间: 2017-11-10 至 2020-03-31
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-17F17797/
• 关键词: エンハンサー同定; リカレントネットワーク; 細胞種特異的エンハンサー; 特徴抽出; Optimization; Genetic Algorithm; DNA; motif finding

The past year has been dedicated to try and identify enhancer regions in the complete human genome by using recurrent neural networks. By taking the whole genome in consideration, and not just some very limited and specific regions, the hope was to reach a more comprehensive understanding of those regions and identify some as of yet unknownenhancers.The results, while significantly better than random, were not as good ashoped. State of the art enhancer identification reaches a success rate of more than 90%, but our results hovered around 60%. While it is definitely a problem that could be tackled in the future, the relatively small dataset available (in opposition to the size of the genome) madeit too difficult for the current machine learning techniques to work: they require both strong ground truth and a big dataset. Sadly the hope that the available data would be enough didn't match reality.The research as since then evolved into a slightly different direction, aiming at being able to identify which enhancer is active in which cell lines. We believe that our experience for studying the motif finding problem using the genetic algorithm would be effective in this direction and thus we will explore this possibility during the remaining term.

过去的一年致力于通过使用递归神经网络来尝试和识别完整人类基因组中的增强子区域。通过考虑整个基因组，而不仅仅是一些非常有限和特定的区域，希望能对这些区域有更全面的了解，并确定一些未知的增强子。结果虽然明显好于随机结果，但也没有预期的那么好。最先进的增强子识别的成功率达到90%以上，但我们的结果徘徊在60%左右。虽然这肯定是一个可以在未来解决的问题，但相对较小的数据集（与基因组的大小相反）使得当前的机器学习技术难以工作：它们需要强大的基础事实和大数据集。遗憾的是，现有数据足够的希望与现实不符。从那时起，研究朝着一个稍微不同的方向发展，旨在能够识别哪些增强子在哪些细胞系中起作用。我们相信，我们的经验，研究基序发现问题，使用遗传算法将是有效的，在这个方向上，因此，我们将探讨这种可能性在剩余的任期。