Session Chair Profile

Giulia C. Kennedy

Biography

Since joining the genomic diagnostics company in 2008, Dr. Kennedy has overseen the development of three commercialized genomic tests that help reduce unnecessary surgeries and healthcare costs by resolving diagnostic uncertainty. The company’s tests are setting new standards in disease diagnosis, where they are changing patient care and obtaining insurance coverage. Through Dr. Kennedy’s scientific leadership, Veracyte continually pushes the boundaries of what is possible in genomic diagnostics, using big data and machine learning to develop tests that answer critical clinical questions. Prior to Veracyte, Dr. Kennedy led the Genomics Collaborations and Genotyping Technology R&D groups at Affymetrix, Inc. Before that, she was a scientific leader for the colon cancer and breast cancer gene discovery efforts at Chiron Corporation, resulting in the identification of oncology markers for therapeutic drug development. Dr. Kennedy was previously a scientist at Millennium Pharmaceuticals, where she implemented genomic and genetic approaches to uncover diabetes susceptibility genes. She holds a Ph.D. degree in biochemistry, and completed postdoctoral training at the University of California at San Francisco in the Biochemistry Department and Hormone Research Institute. Dr. Kennedy has published more than 50 articles in peer-reviewed scientific journals and is a co-inventor on more than 20 patents.

Talk

Clinical Dx Showcase: Bringing in the Machines: When Answering Important Clinical Questions Requires Both Big Data and Big Algorithms
Big data in medicine was traditionally composed of large numbers of patients studied for relatively few variables or features. For example, in the 1950’s, a handful of protein markers for blood groups were measured in half a million people. Even today, current clinical applications mainly report small numbers of gene variants, e.g. cancer panels for therapy selection. Such tests do not require algorithms, as measurements are either positive or negative.

Advances in NGS have created disproportionately huge datasets in feature space and offer the promise of solving difficult clinical questions. However, realizing the potential of these complex datasets requires matching algorithmic interpretation and commercial diagnostic tests that use machine learned algorithms, i.e. “signatures”, are still in the minority.

The diagnosis of patients with idiopathic pulmonary fibrosis is challenged by a multitude of pathology subtypes, variability in biopsy sampling, and cell and tissue heterogeneity. Simply measuring gene mutations lacks diagnostic sensitivity. We have developed and commercialized diagnostic tests with proven clinical utility using large feature sets (x) composed of RNA transcript expression, variant and copy number data and mitochondrial gene expression. Machine learning on this vast feature set allowed us to exploit the richness of NGS data while solving challenging diagnostic dilemmas in thyroid cancer, lung cancer and interstitial lung disease.