Uncovering the epitranscriptome regulatory codes using machine learning

使用机器学习揭示表观转录组调控代码

• 批准号: 10470221
• 负责人: Jianqiu Zhang
• 金额: $ 11.25万
• 依托单位: UNIVERSITY OF TEXAS SAN ANTONIO
• 依托单位国家: 美国
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-09-01 至 2024-08-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10470221
• 关键词: 3' Untranslated Regions; 5' Untranslated Regions; Address; Affect; Algorithms; Area; Bayesian Method; Binding; Biological Process; Code; Data; Detection; Development; Disease; Future; Gene Expression; Gene Expression Regulation; Genes; Goals; High-Throughput Nucleotide Sequencing; Human Herpesvirus 8; Hypermethylation; Knowledge; Lead; Learning; Link; Machine Learning; Malignant Neoplasms; Malignant neoplasm of lung; Mammalian Cell; Mediating; Medical; Messenger RNA; Methods; Methylation; Modeling; Modification; Pathway interactions; Play; Problem Solving; Proteins; RNA-Binding Proteins; Regulation; Research; Resolution; Role; Site; Statistical Distributions; Supervision; Technology; Time; Transcript; Validation; Work; base; cell transformation; convolutional neural network; deep learning; deep learning algorithm; disease phenotype; effective therapy; epitranscriptome; gene function; improved; leukemia; mRNA Stability; machine learning algorithm; malignant breast neoplasm; metaplastic cell transformation; novel; predictive modeling; prevent; tool; transcriptome; tumorigenesis

Project Summary N6-methyladenosine (m6A) is the most abundant methylation widely found in mRNAs of mammalian cells whose function is largely unknown. Recent research has accumulated increasingly strong evidence of m6A's involvement in different diseases such as leukemia, breast cancer, lung cancer, and AIDs. While m6A's close involvement in many diseases is apparent, mechanistic evidence linking m6A alterations to disease phenotypes is mostly missing. Most of the recent research is fueled by the high throughput sequencing technologies such as MeRIP-seq for transcriptome-wide profiling of m6A methylation. However, due to the innate limitations of such technologies, sophisticated machine learning based algorithms are urgently needed to address the problem of detecting m6A sites with high sensitivity and precision, accurate quantification of m6A methylation levels and the prediction of m6A sites differentially affected under disease and normal conditions and the prediction of genes whose expression levels are regulated by m6A. Without bridging these knowledge gaps, it is impossible to made inroads to the problem of finding m6A's role in regulating diseases. To address these issues, our aims in this proposal are 1) Establish a deep learning algorithm for base-resolution m6A site prediction; 2) Establish base- resolution m6A differential site prediction using a hierarchical Bayesian approach; and 3) Determine m6A- mediated genes and functions by Bayesian Negative-Binomial regression. The proposed research will employ deep learning and adversarial learned inference methods and utilize both methylation quantification and sequence information for the first time. Also, it will employ Bayesian graphical model-based methods for combining sequence and methylation level information. It is expected that the developed algorithms will have broad applications in functional study, for which we plan to closely work with our collaborators in applying these algorithms in their research of Kaposi's sarcoma-associated herpesvirus (KSHV), which will lead to the fulfillment our long term goal in the eventual validation and practical medical application of m6A research in the future.

项目概要 N6-甲基腺苷 (m6A) 是哺乳动物细胞 mRNA 中广泛发现的最丰富的甲基化，其 功能很大程度上是未知的。最近的研究积累了越来越有力的 m6A 证据 涉及不同的疾病，如白血病、乳腺癌、肺癌和艾滋病。当 m6A 接近时 明显参与许多疾病，机制证据将 m6A 改变与疾病表型联系起来 大部分缺失。最近的大多数研究都是由高通量测序技术推动的，例如 MeRIP-seq 用于 m6A 甲基化的全转录组分析。然而，由于这种先天的局限性 技术，迫切需要基于复杂机器学习的算法来解决 高灵敏度和高精度检测 m6A 位点，准确定量 m6A 甲基化水平和 预测疾病和正常条件下不同影响的 m6A 位点以及基因预测 其表达水平受 m6A 调节。如果不弥合这些知识差距，就不可能 寻找 m6A 在调节疾病中的作用的问题取得了进展。为了解决这些问题，我们的目标是 建议是1）建立用于碱基分辨率m6A位点预测的深度学习算法； 2) 建立基地- 使用分层贝叶斯方法进行分辨率 m6A 差异位点预测； 3) 确定 m6A- 通过贝叶斯负二项式回归介导基因和功能。拟议的研究将采用 深度学习和对抗性学习推理方法，并利用甲基化定量和 第一次的序列信息。此外，它将采用基于贝叶斯图形模型的方法 结合序列和甲基化水平信息。预计所开发的算法将具有 在功能研究中的广泛应用，我们计划与我们的合作者密切合作来应用这些 他们在卡波西肉瘤相关疱疹病毒（KSHV）研究中使用了算法，这将导致实现 我们的长期目标是未来 m6A 研究的最终验证和实际医学应用。