CfA seminar "Better Data by Design: The Wonderful Yet Frightening World of Post-Genomic Clinical Biomarker Discovery"
Welcome to the CfA seminar by David Broadhurst, Departments of Medicine and Mathematical and Statistical Sciences, University of Alberta, Edmonton, Canada. Sandwiches and refreshments will be served. Please contact Craig Wheelock if you would like to meet up with David Broadhurst during his stay in Stockholm.
About David Broadhurst
Dr. Broadhurst has spent the last 16 years working in Academia with general research interests in Mathematical Modelling, Biostatistics, Machine Learning, Design of Experiments, and Data Visualization.
The main focus of his applied research has been in Metabolomics - the systematic, and data-driven, study of temporal interactions between the compliment of low molecular weight (bio)chemicals (metabolite) abundant within living organisms, tissues and cells. In particular his research focuses on development of workflows that will allow detailed and reproducible investigation of microbial and mammalian metabolomes; therein, to both identify phenotypic biomarkers and further aid the understanding of underlying biological mechanisms for a given organism. He has been involved in multiple large-scale metabolomics-based studies examining human pathology including investigations to identify metabolite biomarkers for two major pregnancy diseases: Preeclampsia and Fetal Growth Restriction.
Since 2011 he has had an appointment as Assistant Professor of Biostatistics at the Faculty of Medicine, University of Alberta, Canada, where he continues to be pursue research in personalized (or precision) medicine involving a wide range of complementary Omic platforms (metabolomics, transcriptomics, microRNA, GWAS, deep sequencing etc.). In 2013, he was appointed to the Faculty of Science (UoA) as an Adjunct Professor in the Department of Mathematical and Statistical Sciences
Summary of his seminar:
The term 'data driven science' is widely used in the post-genomic, systems-biology community, where there has been a history of performing experiments with no clear hypothesis - so called hypothesis-free experiments. It is assumed that by simply collecting arbitrary amounts of high dimensional data regarding the perturbed homeostasis of any given biological system, and applying modern computationally intensive machine learning algorithms, reliable information will simply "drop-out" with very little forethought about the design of such experiments.
Design issues such as: sample size, sample selection, limitations of analytical platforms, accuracy, precision, quality control, appropriate statistical analysis and potential for false discovery, are rarely given detailed consideration. This, combined with a myopic dependence on a small number of very powerful computational methods, together with little understanding of post-hoc statistical analysis, has resulted in a plethora of poor published research that will act only to discredit high throughput science when found to be unrepeatable.
For this seminar, I summarise these problems from a practical perspective, and provide pointers to assist in the improved design and evaluation of biomarker discovery experiments, with the emphasis on clinical metabolomics studies.Contact person: Craig Wheelock