Collaborative Research: MFB: Deciphering RNA-based regulatory logic with interpretable machine learning

RNA转录本是由DNA转录合成的单链核糖核酸产物，包含多个复杂且重叠的密码，这些密码决定了它们的生化过程。RNA剪接和5‘端非编码区调控这两个RNA过程在从DNA到功能RNA和蛋白质产物的基本信息传递中起着关键作用。理解这些基于RNA的密码的调控逻辑对于合理设计生物技术中的RNA转录本是必要的。尽管进行了数十年的遗传学、生物化学和生物信息学研究，但对基于RNA的调控逻辑的理解仍然难以捉摸。最近，“现成”机器学习方法应用于有限的数据集，对潜在的调控逻辑提供了有限的洞察力，阻碍了合理的RNA转录设计。在这个合作项目中，将设计、部署和训练“可设计解释”的机器学习算法，这些算法可以解释它们是如何得出预测的，并将通过实验验证来证明可解释性。该项目将产生更广泛的影响，通过开发可应用于其他生物分子系统的通用实验和机器学习方法，其对生物技术、招聘、参与和受训人员的专业发展的潜在影响，重点支持在科学中代表性不足的不同背景的学生和研究人员，以及为计算机科学和生物学本科生制定课程。这个生物技术的分子基础(MFB)项目专注于两个应用：(1)全面了解剪接密码在确定外显子跳跃中的作用。在RNA剪接过程中，内含子被移除，外显子被连接以形成成熟的RNA转录本。目前还不清楚外显子的序列究竟如何决定它是否会被包括在内。使用一个具有350,000个结构的大规模平行报告分析数据集，并使用可解释的神经网络进行分析，研究人员获得了对剪接密码的多种见解。在这个项目中，实验的RNA结合蛋白结合数据将被纳入，以捕捉序列特征之间的非相加效应，并阐明二级结构的影响。(2)全面了解5‘非编码区在翻译起始和RNA稳定性中的作用。5‘端非编码区序列在调节翻译和稳定性方面起着重要作用，但我们对5’端非编码区的理解还远远不够。在这个项目中，将根据初步分析以及关于microRNA的信息和双向5‘UTR码，设计可解释的神经网络来破译5’UTR码。这个项目将展示可解释的机器学习可以用来破译基于RNA的调控逻辑，这是基础研究的关键一步，对生物技术应用具有直接和普遍的影响。该项目由分子和细胞生物科学部门(MCB)、化学部门(CHE)和信息和智能系统部门(IIS)联合支持。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

RNA transcripts, single-stranded ribonucleic acid products synthesized by transcription of DNA, contain multiple, complex, and overlapping codes that dictate their biochemical processing. Two RNA processes, RNA splicing and 5’UTR regulation, play key roles in the fundamental transfer of information from DNA to functional RNA and protein products. Understanding the regulatory logic of these RNA-based codes is required for the rational design of RNA transcripts in biotechnology. Despite decades of genetics, biochemistry, and bioinformatics research, understanding RNA-based regulatory logic remains elusive. Recent applications of “off-the-shelf” machine learning methods to limited datasets have provided limited insights into the underlying regulatory logic, hindering rational RNA transcript design. In this collaborative project, “interpretable-by-design” machine learning algorithms that can explain how they arrive at their predictions will be designed, deployed, and trained on massively parallel reporter assays and interpretability will be demonstrated by experimental validation. The project will have broader impacts through the development of generalizable experimental and machine learning approaches that can be applied to other biomolecular systems, its potential impact for biotechnology, recruitment, participation, and professional development for trainees, with an emphasis on supporting students and researchers from diverse backgrounds underrepresented in the sciences, and development of curricula for undergraduate students in computer science and biology. This Molecular Foundations of Biotechnology (MFB) project is focused on two applications: (1) A comprehensive understanding of the splicing code in determining exon skipping. During RNA splicing, introns are removed, and exons are ligated to form the mature RNA transcript. It is currently unknown how exactly an exon’s sequence determines whether it would be included or not. Using a massively parallel reporter assay dataset with 350,000 constructs and analyzed using an interpretable neural network the investigators have derived multiple insights into the splicing code. In this project, experimental RNA binding protein binding data will be incorporated to capture non-additive effects among sequence features and elucidate the effect of secondary structure. (2) A comprehensive understanding of the role of the 5’UTR code in translation initiation and RNA stability. The 5’UTR sequence plays important roles in regulating translation and stability, yet our understanding of this 5’UTR code is far from complete. In this project, interpretable neural networks will be designed to decipher the 5’UTR code, based on preliminary analysis, as well as information about microRNAs, and bidirectional 5’UTR scanning. This project will demonstrate that interpretable machine learning can be used to decipher RNA-based regulatory logic, a critical step forward for basic research with direct and generalizable implications for biotechnology applications.This project is jointly supported by the Division of Molecular and Cellular Biosciences (MCB), the Division of Chemistry (CHE), and the Division of Information and Intelligent Systems (IIS).This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.