Progress in medical science has always been driven by high quality data. A longstanding barrier to progress, both in clinical settings and research trials, has been the fundamental difficulty of accurately measuring the human phenotype, including but not limited to behavioral patterns, social interactions, physical mobility, gross motor activity, and speech production. Smartphones are now ubiquitous and can be harnessed to offer medicine a wealth of data on disease phenotypes. Subjects can be instrumented using dedicated scientific apps that generate continuous streams of behavioral data. This approach has enormous potential, but data alone are not enough: the data need to be coupled with appropriate statistical learning techniques–techniques that are specific to the given domain and driven by specific scientific questions, in order to fully leverage their potential. Only then can the data be transformed into clinically valid and useful information, the kind of information that can transform research and discovery and, ultimately, patient health.
We have coined the term digital phenotyping to refer to the moment-by-moment quantification of the individual-level human phenotype, in situ, using data from personal digital devices, such as smartphones. Human behavior is context specific, which is why characterizing behavior in situ–life as lived and experienced by people in their everyday surroundings–is critical. Our understanding of the human genome has been revolutionized in the past two decades, but there are at present many fields in medicine where some of the most important challenges lie in characterizing the associated disease phenotypes. This is where digital phenotyping comes in. (Read more about smartphone based digital phenotyping here.)
We have developed the Beiwe Research Platform consisting of a study portal, smartphone app, backend database, and data modeling and analysis tools. Unlike the large majority of commercially available smartphone applications, Beiwe is intended for biomedical research, and the most important aspect of the smartphone app is the collection of raw sensor and phone use data. Our primary activity in this area is the development of statistical learning methods, and their implementation in software, to analyze and model data collected by the Beiwe app. Our initial application areas of digital phenotyping are in psychiatric, neurological, and surgical patients, where we aim to generate new outcome measures and social and behavioral biomarkers. Our work in Digital Phenotyping is supported primarily by an NIH Director’s New Innovator Award made to Dr. Onnela (see HSPH Press Release).
Statistical Network Science
Many systems of scientific and societal interest consist of a large number of interacting components. The structure of these systems can be represented as networks, where network nodes represent the components and network edges the interactions between the components. Network analysis is used to study how pathogens, behaviors and information spread in social networks, having important implications for our understanding of epidemics and the planning of effective interventions. In a biological context, at a molecular level, network analysis can be applied to gene regulation networks, signal transduction networks, protein interaction networks, and more.
Network science is an interdisciplinary field that draws heavily from mathematics, statistical physics, computer science, social sciences, and many others. Two major paradigms to the study of networked systems are the physics-based approach and the statistics-based approach. The former involves modeling network phenomena whereas the latter involves modeling data that arrive in the form of a network. In terms of methodology, much of our work focuses on bridging some of the gap that exists between these two approaches, hence the term statistical network science.
We have several methodological and applied projects going on at any given time. For example, in one line of work, we attempt to bring mechanistic network models from physics to the realm of statistical inference. We are also involved in number of applied projects that make use of network methods to study a range of biomedical and public health problems. Among others, our ongoing projects deal with structural properties of large-scale social networks, network inference, healthcare outcomes and coordination of medical care, epidemics on networks, and HIV/AIDS cluster randomized trials. We also make use of call detail records, or CDRs, to study human social networks and mobility patterns at scale.