Systems biology and machine learning analytics for single-cell RNA sequencing and skin microbiome data

Abstract

Rapid advancements in next-generation sequencing (NGS) technologies have exponentially accelerated the access of large-scale, high-resolution, and reliable data and revolutionized medical and health research. In this thesis, systems biology paradigm coupled with machine learning models have been applied for studies of single-cell RNA-sequencing (scRNA-seq) and skin microbiome in psoriasis, two representative areas of research that benefit from the development of NGS. scRNA-seq is one of the latest sequencing technologies that enables gene expression profiling at the single-cell resolution, which deepens the understanding of living systems by orchestrating the function of cell, the basic unit of life. The high-dimensional, noisy, and sparse scRNA-seq data increases the difficulty in analysis. Despite previous studies have utilized various machine learning based computational methods to learn hidden patterns in scRNA-seq data for identifying cellular heterogeneity, challenges still remain. This call for additional questions to be addressed. For example, whether partitioned cell clusters are biologically meaningful and what are the heterogeneous functions of different cell clusters. In this thesis, a pathway-based machine learning analytic approach to cluster cell populations, identify discriminative functional pathways related to cellular heterogeneity and construct gene-gene interactions networks with the highlight of ‘hub’ genes bridging the cross-talk of different pathways has been proposed. As genes work interactively at the pathway level, the proposed approach therefore improves the credibility of cell clustering and enhances the interpretation of biological functions of different cell populations. Advances in NGS also allows the characterization of the skin microbiome, a community of diverse microorganisms residing on the skin. Previous studies have demonstrated the ability of skin microbiome to educate host immune system and an association between the skin microbiome and psoriasis. However, most conducted studies only focused on the bacterial community of skin microbiome and its taxonomic composition. Using the whole genome metagenomic shotgun sequencing, this thesis provides a comprehensive investigation of skin microbiome that characterizes both taxonomic and functional profiles of bacteria and less explored bacteriophage components of the skin microbiome in psoriasis patients and in family controls. Altered compositions of both bacterial and bacteriophage communities in psoriatic skin compared to healthy skin have been observed. The capacity to suppress host bacteria by bacteriophage has been revealed from both taxonomic abundance and functional protein levels. Moreover, the temporal dynamics of skin microbiome in patients with psoriasis compared to controls has been examined. Key microbes associated with the change of skin status in psoriasis have been identified and the prediction of skin status using machine learning has been achieved. This thesis reinforces the important role of dynamically stable and balanced skin microbiome in maintaining skin homeostasis and serves as an inspiration to future probiotic therapeutics. This thesis has provided novel and systematic analytic framework for scRNA-seq and shed new light on the role of skin microbiome in psoriasis. By harnessing the holistic systems biology paradigm along with machine learning to facilitate the understanding, prediction and monitor of physiological processes and disease progression, ultimately this thesis aims to promote a better practice of precision and personalized medicine.