Abstract

Pathology (laboratory medicine) is central to modern healthcare, and as such generates vast amounts of data from populations, including profiles of routine serum/blood investigations, and results from specialised, second tier laboratory tests. Routine testing alone is comprehensive, covering the function of physiological systems in health and disease. With millions of results generated in Australia each year, a huge data repository reflecting human physiology, biochemistry and pathology is available to inform population and biomedical research.

This presentation provides an example of how the machine-learning (pattern recognition) algorithms of recursive partitioning (trees and forests), and support vector machines, are applied to extend diagnostic pathology as a research resource, and via further development, answer complex health questions. The example explored is infection by hepatitis B virus (HBV), and how pattern recognition via machine learning allows simpler prediction of infection via routine data, as well as the further development of pattern recognition strategies to monitor HBV persistence in patients.

About Brett

Brett has a laboratory background, from diagnostic pathology through to research into virus-host interaction and pathogenesis. Post-doctoral experience included molecular virology and mucosal vaccine development. Several years were devoted to lecturing molecular biology, genetics and medical science, during which time fundamental research endeavours evolved on the immuno-pathogenesis of Ross River virus (RRV) infection.

Experimental laboratory-based research into infectious disease and pathology has been replaced by in silico methods, with disease questions addressed via machine-learning interrogation of large pathology datasets, as well as via data collected through volunteer studies. Current research programmes involve the elucidation of biomarker patterns for ME/CFS, investigations of pathology data quality, and augmented predictive efficacy to enhance laboratory diagnoses.