Developing Ribolog: A toolbox for comprehensive analysis of ribosome profiling data

摘要 核糖体图谱技术为基因组水平的翻译调控提供了定量的见解， 在几个重要的生物过程中发挥关键作用的机制，从胚胎发育到 致癌。尽管核糖体图谱数据分析方法取得了进展，但仍存在一些挑战 需要解决的问题包括小样本量的测试和由于核糖体停滞造成的读取计数偏差。我 建议开发一种名为“Ribolog”的基于Logistic回归的方法来模拟核糖体图谱数据 哪些单独的测序读数是观察单位，翻译效率被计算为赔率 观察“RPF”与“RNA”是这样写的。Logistic回归模型具有几个明显的优势 基于RNA-seq和ribo-seq读取计数的负二项模型的方法：(I)它既不假设 均值和方差相等，也不需要估计离散度。(Ii)它有更高的统计数据 比基于计数的方法更强大，因为在该模型中，统计样本大小等于读取次数， 而不是复制的数量。(3)它适用于每个条件下的单个样本(未复制的数据集)；因此， 适用于临床或单细胞数据。(Iv)它很容易适应合成加剂的实验 标准。(V)在复制的数据集上，它允许对新颖的数据进行经验意义测试和计算 信息丰富的质量控制措施。(Vi)它可以容纳涉及多个样本的复杂实验设计 和协变量在一个模型中；并且不限于成对比较。我们的初步结果适用于 对包括两个非转移细胞系和两个相应转移细胞系的数据集的Ribolog表明 这种方法确实非常强大，在生物复制中具有80%-90%的重复性。此外，我们 提供失速偏差校正、荟萃分析、模型选择、实验设计和质量模块 控制力。将Ribolog与其他分析方法相结合--其中一些是我们实验室以前开发的--我们 构建核糖体序列数据与RNA-SEQ、tRNA图谱、遗传变异、 MiRNA、密码子最优化等，以确定翻译动态的驱动原因并为下一步做出贡献 多层基因-表型图谱的生成。该方法将在R中实现并生成 以开放获取包的形式向科学界提供。鉴于我在统计学方面的专业知识，我的 继续接受实验生物学方面的培训，获取最先进的数据集，并获得多个实验室的支持 在翻译和更广泛的基因组主题的计算和实验研究方面拥有专业知识 技术，我处于解决这个问题的独特位置。除了提供对 翻译控制的生物学，并使社区受益，这个项目将使我能够延长我的培训 在一些与我的职业发展最相关的激动人心的领域，作为一名成功的 独立的学术研究科学家。

ABSTRACT Ribosome profiling technology provides quantitative insights into translational regulation at a genomic scale, a mechanism that plays a crucial role in several important biological processes from embryonic development to carcinogenesis. Despite advances in ribosome profiling data analysis methods, a number of challenges remain to be addressed including tests with small sample sizes and read count biases due to ribosome stalling. I propose to develop a logistic-regression-based method called “Ribolog” to model ribosome profiling data in which individual sequencing reads are units of observation and translation efficiency is calculated as the odds of observing “RPF” vs. “RNA” reads. The logistic regression model has several distinct advantages over the methods based on negative binomial modeling of RNA-seq and Ribo-seq read counts: (i) It neither assumes equality of mean and variance nor does it require estimation of dispersion. (ii) It has much higher statistical power than count-based methods because in this model, statistical sample size equals the number of reads, not the number of replicates. (iii) It works with single sample per condition (unreplicated datasets); therefore, it is applicable to clinical or single cell data. (iv) It is easily adaptable for experiments with synthetic spike-in standards. (v) In replicated datasets, it enables empirical significance testing and calculation of novel informative QC measures. (vi) It can accommodate complex experimental designs involving multiple samples and covariates in one model; and is not limited to pairwise comparisons. Our preliminary results applying Ribolog to a dataset comprising two non-metastatic and two corresponding metastatic cell lines indicate that this method is indeed highly powerful and 80-90% reproducible among biological replicates. Additionally, we provide modules for stalling bias correction, meta-analysis, model selection, experimental design and quality control. Combining Ribolog with other analytical methods – some developed previously in our lab – we construct a multiomic framework to integrate Ribo-seq data with RNA-seq, tRNA profiling, genetic variation, miRNA, codon optimality etc. to identify the driving causes of translation dynamics and contribute to the next generation of multi-layered genotype-to-phenotype maps. The method will be implemented in R and made available to the scientific community as an open-access package. Given my expertise in statistics, my continued training in experimental biology, access to state-of-the-art datasets, and support from multiple labs with expertise in computational and experimental studies of translation and broader genomic topics and technologies, I am uniquely situated to tackle this problem. In addition to providing novel insights into the biology of translational control and benefiting the community, this project will enable me to extend my training in a number of exciting areas that are most relevant to my career development as a successful and independent academic research scientist.