Organizing Medical and Genomic Knowledge for Precision Medicine
The widespread adoption of Electronic Health Records (EHRs) has enabled the collection of patient health information at an unprecedented granularity and scale. Despite the vast scale of EHR data collection efforts, however, we’ve seen relatively little in the way of data-driven innovation affecting clinical care on a comparable scale. The potential is great, but there remain challenges to overcome. As many of us have observed over the past few years, the stumbling blocks have been commercial and political, rather than technical. Many health care providers are reluctant to share data across affiliation lines to other health systems, and the fragmentation of data means that machine learning researchers cannot easily access the vast amount and array of rich real-world clinical data necessary to build intelligent systems.
Imagine a world where longitudinal health records on tens of millions of people were available to the best data scientists and machine learning researchers. Imagine there were access to extremely large and complex data sets, where the opportunity existed through machine learning to discover associations and understanding of patterns and trends within the data; and more, uncovering hidden patterns that might never have been evident in the past. This potential can fundamentally transform how care is delivered and reimbursed; transforming how treatments are discovered.
At Flow Health, our strategy has been — ‘build it and they [data] will come’.
Over the past several years, we’ve been building a unique data platform that unifies structured and unstructured health data from multiple sources into a central person-centered data store and integrates data into a graph-based data model with dynamic ontologies, where data is categorized and organized around the people, places, things and events that it relates to.
This is the foundation upon which we can create human-level intelligent systems.
We envision an intelligent platform that can generate data-driven knowledge from the co-occurrence of relationships in health data. There are huge leaps from data to information to knowledge to automated reasoning. Through machine learning, we can derive patterns in disease progression in order to predict prognosis, look at outcomes data from treatment pathways to suggest treatment plans at the genetic level — and this is just the beginning.
By applying unsupervised deep learning to massive data, we expect to identify new biomarkers and precise phenotypes that can serve as the foundation for building personalized diagnostic and treatment pathways. With the advances in deep learning, this will enable better predictions and automated reasoning to personalize medical decisionmaking.
Andrew Ng, Chief Scientist at Baidu Research and the founder of the Google Brain project, uses the analogy of building a rocket ship when thinking about creating deep learning. In Andrew Ng’s words, “A rocket ship is a giant engine together with a ton of fuel. Both need to be really big. If you have a lot of fuel and a tiny engine, you won’t get off the ground. If you have a huge engine and a tiny amount of fuel, you can lift up, but you probably won’t make it to orbit. So you need a big engine and a lot of fuel.”
The tools to build the big rocket engine exist; high-performance cloud computing, distributed data processing systems and deep learning neural networks. What is needed is fuel — the data. This is what we’ve been focused on: gathering the fuel necessary to take the rocket ship into the next frontier of healthcare — where no one has gone before. We’ve overcome a major challenge and have access to the needed fuel.
We’re thrilled to have recently formed a ground-breaking partnership that will provide the necessary data to build a medical knowledge graph that can understand the relationships between symptoms and conditions, treatments and outcomes, medications and side effects, phenotypes and genotypes, and more; training deep learning algorithms to make precision medicine a reality. Stay tuned for further details.