Computational Challenges of Mass Phenotyping

One of the primary challenges in making sense the dramatic increase in human genotype data is finding suitable phenotype information for correlational analyses. While the price of genotyping has fallen dramatically and promises to continue to decrease, the cost of generating the phenotypes necessary to take advantage of this data has held steady or even increased. Until recently, human phenotype data was primarily derived from assays or measurements made in clinical or research laboratories. However, laboratory phenotyping is expensive and low-throughput. Recently, a variety of promising alternatives have arisen that can provide important new information at greatly reduced costs. However, the nature, extent and complexity of the data produced involve significant new computational challenges. This workshop will begin with an introduction to some of the new modalities, which include: automated abstraction of information from electronic medical records, data streams from medical instruments (e.g. in an intensive care unit) and implanted devices (e.g. cardiac assist devices), data produces by patient social networks, and data from a new generation of inexpensive wearable sensors measuring everything from physical activity to blood glucose. Most of these new sources of phenotypic data are secondary to some other purpose. Patient records are generated to support clinical care and payment for medical services. Patient social networking sites support patients emotionally and provide peer counseling. Implantable medical devices produce data streams that meet manufacturers' or caregiver requirements. Wearable sensors satisfy personal curiosity or monitor disease progress. Each of these also produces valuable information for genotype correlations. We will focus on defining the computational challenges arise in the collection, storage, processing, analysis and, especially, in the useful integration of these many new sources of phenotype data into derivatives that facilitate scientifically or medically valuable correlations with genotype. Computational challenges arise due to the diverse nature of the types of data that characterize human phenotypes, the fact that most phenotyping is a secondary use of data produced for other purposes, and the need to integrate, abstract and summarize data in ways that are likely to show correlations with genotype. There are also bioethical challenges in data sharing, anonymization, openness / privacy, consent, and related topics where computational methods might help address other concerns. The new sources of phenotypic information produce data at radically different time scales and granularities. Modern medical instruments can produce data streams at 50Hz or greater sampling frequencies for days at a time. Patient social networking users …