Inferring Respiratory Disease Biology from Geolocation Data

Biological fitness quantifies the efficiency and selective advantage of pathogens and hosts in their bilateral interaction. Key questions—such as how much more infectious an emerging variant is compared with its predecessor, or how much protection vaccination offers relative to no vaccination—require fitness to be measured systematically, in real time, and ideally beyond controlled laboratory settings. We propose an approach that infers biological fitness from mostly non-biological data on infection dynamics and contact levels in a population. Because contact levels are expected to predict infection levels under stable biological conditions, systematic deviations between these trends can indicate changes in the underlying processes of transmission or immunity. Using infection surveillance data and GPS co-location information as a proxy for social contact patterns, we apply Bayesian modeling to detect and quantify biological shifts throughout the SARS-CoV-2 pandemic in Germany. We identify substantial regional variation in both the timing and magnitude of fitness changes. These shifts align with the emergence of major variants and the accumulation of population immunity; across Germany, the Alpha, Delta, and Omicron variants were 29 %, 63 %, and 108 % more transmissible than wild-type SARS-CoV-2, respectively. Early natural infection, primary vaccination, and booster vaccination increased population-level immunity by 13 %, 94 %, and 114 % relative to the pre-pandemic naïve state. Our approach provides an immediately deployable framework for detecting biological change during emerging epidemics, given that relevant behavioral and surveillance data are collected in real time. It is particularly valuable for low-resource settings where direct biological measurements may be limited. Significance Statement Biological changes in pathogen transmission remain difficult to measure in real time during acute epidemic and pandemic situations. Here, we showcase a method to extract biological insights from Bayesian modeling of anonymized mobile phone-based GPS-derived contact data and anonymized infection data. We estimate changes in transmissibility associated with virus evolution and vaccination and find agreement with epidemiological, virological and immunological evidence. Increasingly available mobile phone data makes our approach more easily amenable for use in real time. It enables near real-time detection of biological changes in ongoing (e.g. Influenza) and future epidemics and may be especially valuable in low-resource settings.