A computational framework to unravel gene regulatory mechanisms using single-cell omics data

识别与特定环境或表型结果相关的转录调控机制是一个重要的生物学问题。转录调控网络（trn）代表了转录因子对基因的调控作用，已被证明是描述不同生物学背景下基因表达程序的有用模型。trn通常是从跨许多样本的大量基因表达数据中计算构建的，单独或与其他数据类型相结合。然而，由于大量基因组数据集代表了样本中“所有”细胞的“平均”概况，基于这些数据类型重建的trn对于研究在单个细胞分辨率下效果最好的过程或样本中包含异质细胞的场景的价值有限。因此，对这些过程及其trn的研究应该只使用参与这些过程的细胞的单细胞组学图谱进行。然而，目前使用单细胞数据重建TRN的方法通常与基于大量数据的方法相对应，因此忽略了使用这些新技术数据时存在的挑战和机遇。此外，这些方法对不同样品的表型标签是不可知的，不能确定对感兴趣的表型变异负责的调节机制。我的计划是开发一个统一的计算框架，允许在包含（潜在的）异质细胞的不同样本中，使用与特定生物表型相关的多组学数据集，以单细胞分辨率重建因果trn。这个计算框架将基于我们将开发的新型机器学习和图挖掘算法来构建。作为我研究计划的一部分，这个框架和它的计算工具将在生物信息学和计算基因组生物学领域产生重大影响，为识别导致表型结果变异的调节机制提供先进的新方法。他们将揭开新的生物学机制，并将确定与表型相关的新的生物标志物。此外，它们可能揭示操纵生物系统（如细胞、疾病、生物体）表型特性的新方法：例如，提高作物对害虫的抵抗力，或开发治疗神经退行性疾病或癌症的新方法。这些工具将作为用户友好的公开软件来实现，也将使实验生物学家（最终用户）能够分析他们的数据，从中提取新的见解，并为他们的实验确定新的目标。最后，我的项目将对培养下一代的计算生物学家和生物信息学家产生重大影响，包括本科生、研究生（硕士和博士）和博士后。

Identifying transcriptional regulatory mechanisms related to a specific context or a phenotypic outcome is an important biological problem. The transcriptional regulatory networks (TRNs), which represent the regulatory effects of transcription factors on the genes, have proven to be a useful model for describing gene expression programs in different biological contexts. TRNs are usually constructed computationally from bulk gene expression data across many samples, alone or in combination with other data types. However, since bulk omic datasets represent the 'average' profile of 'all' the cells in a sample, TRNs reconstructed based on these data types have limited value to study processes whose effects are best observed at a single cell resolution or scenarios in which samples contain heterogeneous cells. Consequently, the study of these processes and their TRNs should be performed using single-cell omics profiles of only cells involved in these processes. However, current methods for TRN reconstruction using single-cell data are usually counterparts of methods based on bulk data, and as such ignore challenges and opportunities that exist when using data from these new technologies. Moreover, these methods are agnostic to the phenotypic labels of different samples and cannot identify regulatory mechanisms that are responsible for the variation in a phenotype of interest. My plan is to develop a unified computational framework that will allow reconstruction of causal TRNs, across different samples containing (potentially) heterogeneous cells, at a single-cell resolution using multiomics datasets, related to a specific biological phenotype. This computational framework will be constructed based on novel machine learning and graph mining algorithms, which we will develop. This framework and its computational tools, developed as part of my research program, will have a significant impact in the fields of bioinformatics and computational genome biology by providing advanced novel methods for identification of regulatory mechanisms responsible for variation in a phenotypic outcome. They will unravel new biological mechanisms and will identify novel biomarkers related to a phenotype. In addition, they may reveal new ways of manipulating the phenotypic properties of a biological system (e.g. a cell, a disease, an organism): for example to improve the resistance of crops to pests or to develop new treatments for neurodegenerative diseases or cancer. These tools, which will be implemented as user-friendly publicly available software, will also enable experimental biologists (the end users) to analyze their data, extract new insights from it, and identify novel targets for their experiments. Finally, my program will have a significant impact by training the next generation of computational biologists and bioinformaticians at various levels of undergraduate, graduate (Masters and PhD) and postdoctoral.