Data mining for systems medicine and spectroscopic profiling: methods and applications

Abstract

With the advent of analytical technologies that quantify the biological and chemical compartments of health and disease in a high throughput manner; there has been a pressing need for computational pipelines to harness the power of these wealth of data. The focus of the thesis was on developing statistical and computational methods and pipelines to pre-process spectroscopic profiling data and to analyse and integrate omics data with the aim of disease phenotyping.

We showed that systems medicine with the aim of stratification of patients is a progressively multidisciplinary that can benefit from computational and statistical methods drawn from various fields such as computer science, statistics and information theory.

It has been increasingly common to acquire multiple omics data from multiple tissue and bio-fluid to understand complex diseases with heterogeneous response to treatments. The goal of such comprehensive studies is to gain systematic insights into the disease mechanism through the lens of multiple omics data sets. An impediment to reach the considerable potential of these technologies is the lack of computational and statistical methods and pipelines to process the mountain of data that they generate. Each data set depending on its underlying technology posses different challenges.

Some require robust per-processing techniques for improved information gain and some passed this phase and require to be coupled with complementary data for improved knowledge discovery. An example of the former is the mass spectrometry imaging and an example of the later is transcrimptomics. In the thesis we investigated pre-processing methods for mass spectrometry imaging data. A hierarchical clustering methods is optimised to stand the challenge of big MSI data set. A spatial statistical pipeline was designed and implemented for MSI to minimise information loss during pre-processing. A challenge in systems medicine is to integrate multiple omics data sources to provide a comprehensive insight into the disease mechanism This thesis proposed a multi-layer pipeline that incorporated methods from bioinformatics, statistics and machine learning fields to successfully integrate multi-omics data and generated clinically relevant subphenotypes for sever asthma.