Pattern Recognition for Complex Heterogeneous Time-Series Data: An Analysis of Microbial Community Dynamics

Microbial life is the most wide-spread and the most abundant life form on earth. They exist in complex and diverse communities in environments from the deep ocean trenches to Himalayan snowfields. Microbial life is essential for other forms of life as well. Scientific studies of microbial activity include diverse communities such as plant root microbiome, insect gut microbiome and human skin microbiome. In the human body alone, the number of microbial life forms supersedes the number of human body cells. Hence it is essential to understand microbial community dynamics. With the advent of 16S rRNA sequencing, we have access to a plethora of data on the microbiome, warrantying a shift from in-vitro analysis to in-silico analysis. This thesis focuses on challenges in analysing microbial community dynamics through complex, heterogeneous and temporal data. Firstly, we look at the mathematical modelling of microbial community dynamics and inference of microbial interaction networks by analysing longitudinal sequencing data. We look at the problem with the aims of minimising the assumptions involved and improving the accuracy of the inferred interaction networks. Secondly, we explore the temporally dynamic nature of microbial interaction networks. We look at the fallacies of static microbial interaction networks and approaches suitable for modelling temporally dynamic microbial interaction networks. Thirdly, we study multiple temporal microbial datasets from similar environments to understand macro and micro patterns apparent in these communities. We explore the individuality and conformity of microbial communities through visualisation techniques. Finally, we explore the possibility and challenges in representing heterogeneous microbial temporal activity in unique signatures. In summary, in this work, we have explored various aspects of complex, heterogeneous and time-series data through microbial temporal abundance datasets and have enhanced the knowledge about these complex and diverse communities through a pattern recognition approach