Preview

PDF, English - main document Download (12MB) | Terms of use

Abstract

Single-cell multimodal genomics involves simultaneous measurement of multiple types of molecular data, such as gene expression, epigenetic marks and protein abundance, in individual cells. This allows for a comprehensive and nuanced understanding of the molecular basis of cellular identity and function. The large volume of data generated by single-cell multimodal genomics experiments requires specialised methods and tools for handling, storing, and analysing it. This work provides contributions on multiple levels. First, it introduces a single-cell multimodal data standard — MuData — designed to facilitate the handling, storage and exchange of multimodal data. MuData provides interfaces that enable transparent access to multimodal annotations as well as data from individual modalities. This data structure has formed the foundation for the multimodal integration framework, which enables complex and composable workflows that can be naturally integrated with existing omics-specific analysis approaches. Joint analysis of multimodal data can be performed using integration methods. In order to enable integration of single-cell data, an improved multi-omics factor analysis model (MOFA+) has been designed and implemented building on the canonical dimensionality reduction approach for multi-omics integration. Inferring later factors that explain variation across multiple modalities of the data, MOFA+ enables the modelling of latent factors with cell group-specific patterns of activity. MOFA+ model has been implemented as part of the respective multi-omics integration framework, and its utility has been extended by software solutions that facilitate interactive model exploration and interpretation. The newly improved model for multi-omics integration of single cells has been applied to the study of gene expression signatures upon targeted gene activation. In a dataset featuring targeted activation of candidate regulators of zygotic genome activation (ZGA) — a crucial transcriptional event in early embryonic development, — modelling expression of both coding and non-coding loci with MOFA+ allowed to rank genes by their potency to activate a ZGA-like transcriptional response. With identification of Patz1, Dppa2 and Smarca5 as potent inducers of ZGA-like transcription in mouse embryonic stem cells, these findings have contributed to the understanding of molecular mechanisms behind ZGA and laid the foundation for future research of ZGA in vivo. In summary, this work’s contributions include the development of data handling and integration methods as well as new biological insights that arose from applying these methods to studying gene expression regulation in early development. This highlights how single-cell multimodal genomics can aid to generate valuable insights into complex biological systems.