JUSTIN GUINNEY, PhD
VP, Computational Oncology, Sage Bionetworks
Affiliate Associate Professor, University of Washington
Director, DREAM Challenges
Washington, USA

Biography (web)

The Model-to-Data Paradigm: Overcoming Data Access Barriers in Biomedical Competitions

Data competitions often rely on the physical distribution of data to challenge participants, a significant limitation given that much data is proprietary, sensitive, and often non-shareable. To address this, the DREAM Challenges has advanced a challenge framework called model-to-data (M2D), requiring participants to submit re-runnable algorithms instead of model predictions. The DREAM organization has successfully completed multiple M2D-based challenges, and is expanding this approach to unlock highly sensitive and non-distributable human data for use in biomedical data challenges.

The EHR DREAM Challenges is the first M2D data utilizing EHR data. We are asking participants to predict patient mortality within six months of their last hospital visit, using data from the University of Washington Medical System enterprise data warehouse (2009-2019) with over 1.3 million patients and 22 million visits. Given the highly sensitive nature of EHR data and risks of re-identifiability, the M2D approach is being used to ensure data privacy and protections.

Prior to launching the EHR Challenge, we completed a feasibility study and developed 3 models: demographic information; demographic information and 4 common chronic diseases, and demographic information and the top 20 indications. Model performance using area under the receiver-operating-curve was 0.682, 0.794, and 0.723, respectively. This demonstrated technical robustness of the challenge architecture, and the ability to generate and evaluate predictive algorithms in a secure manner. The EHR Challenge was launched in September 2019, and will close in January 2020.

- top -

JOYCE C. HO, PhD
Assistant Professor, Computer Science Department
Emory University
Georgia, USA

CV (.pdf)

Developing an Evidence Matching Framework Using Web-based Medical Literature

Researchers are discovering new disease subgroups from secondary analyses of electronic health records. However, such subgroups need to be validated or aligned with current literature. We developed a scalable framework that produces evidence sets (or relevant articles) using a large corpus of online medical literature. In this talk, I will discuss some of the challenges associated with term representation and mining biomedical text. I will also present a case study of our framework to validate EHR-based phenotypes.

- top -

KIRK E. JORDAN, Ph.D.
IBM Distinguished Engineer
Data Centric Solutions
IBM T.J. Watson Research
and
Chief Science Officer
IBM Research UK

Biography: web

Intelligent Simulations - Incorporating AI into Computational Simulations

We are at an inflexion point in the way we are computing today and the way we will compute tomorrow.  While computer systems will consist of heterogeneous components adding to complexity we need to make high end systems much more accessible.   Combine this machine complexity with the fact that more compute is required to tackle the ever increasing complexity of the problems and data that are driving science investigations especially in the life sciences.  How will we address this ever increasing complexity.  IBM Research is looking at the Future of Computing through a new lens which consists of bits, neurons and qubits.  In this talk I will describe how IBM Research is looking at the future of computing and I will outline how we are increasing the use of Artificial Intelligence (AI) and Machine Learning (ML) to address some of the complexity while enabling  us to tackle the interesting science questions we and our collaborators are currently investigating with some examples in the Life Sciences.

- top -

HEINRICH RÖDER, PhD
Biodesix, Inc.
Colorado, USA

Biography (web)