Abstract

The health status of an organism modulates the dynamic and complex interplay of biochemical species that make-up the body and fluids of the organism. As such, these biological fluids are routinely used for diagnostic testing, yet they are often not used to their full potential. For instance, amniotic fluid (AF), the fluid that surrounds the fetus during gestation, is collected primarily for genetic testing from women with identified risk factors. The AF proteome and/or metabolome are seldom considered and represent a largely untapped wealth of relevant clinical information. Extensive, multi-analyte data can be collected from biological samples with modern analytical instrumentation. However, sophisticated data preprocessing and analysis (i.e. chemometrics) are required to reveal the relationships between the biochemical signals and the health status. This thesis seeks to demonstrate that untargeted biomarker discovery strategies can be efficiently applied to the task of finding novel biomarkers and complement the traditional hypothesis driven approaches.
In the work underlying this thesis, a chemometric data analysis strategy was developed to search for biomarkers in capillary electrophoresis (CE) separations data. The absorbance data from amniotic fluid samples (n=107) collected at 15 weeks gestation, at 195 +/- 4 nm, was normalized, time aligned with Correlation Optimized Warping and reduced to a smaller number of variables by Haar transformation. The reduced data was then classified into normal or abnormal health classes by using a Bayes classifier algorithm.
The chemometric data analysis was first employed to find biomarkers of gestational diabetes mellitus (GDM) and revealed that human serum albumin (HSA) could predict the early onset of disease. The same approach was successfully used to identify cases of large-for-gestational age (LGA) with the same AF CE-UV data. It was also employed for the classification of embryos with high and low reproductive potential using in vitro fertilization (IVF) culture media analyzed by CE-UV.
Overall, a chemometric method was developed to perform untargeted biomarker discovery in biological samples and provide new means to detect GDM pregnancies, LGA neonates and viable embryos in IVF. The method was successful at identifying biomarkers of interest and showed high flexibility and transferability to other biological fluids.