Grant: Discovery Grants Program - Individual

As demonstrated during the COVID-19 pandemic, wastewater sampling can be used to monitor viruses and bacteria. Public health agencies have recognized the potential of wastewater sampling beyond the SARS-CoV-2 virus, particularly due to its unbiased and non-invasive nature. However, wastewater samples are only able to collect small fragments of the original genome sequence, and thus we cannot simply compare the genetic sequences sampled from wastewater to known definitions of "variants" (i.e., the classification system that delineates which genetic sequences belong to different groups). Current methods for detection and estimation of proportions of variants in wastewater are limited to analyzing a single sample and require a fixed definition of the variants (based on a list of mutations relative to a reference sequence) as obtained from global clinical samples (which may differ from the local circulating variants). I will improve on these methods using machine learning techniques to choose which variants to search for, incorporate the evolutionary structure of the variants, and allow for a more robust estimate of the uncertainty in the estimates. I will develop a model that incorporates information from the same wastewater treatment plant (WWTP) on previous days (temporal information) and WWTPs within the same city (spatial information). It is unlikely that variant that is present in high proportion one day will drop to 0 the next day, and proportions should be similar for WWTPs that are close to each other in space. An alternative to directly modelling the change in mutation lists is to use unsupervised clustering algorithms. This approach involves looking at numerous samples and identifying clusters of mutations (i.e., variants) that tend to appear together. Existing machine learning models for finding clusters that appear together will be extended to accommodate the coverage of each mutation, a changing number of clusters, and clusters that have combined (i.e., recombinant viruses). There is spatio-temporal information in the samples which can be incorporated by constraining the estimates of clusters in samples that are "close" (in space or time) to be more similar than further away samples and allowing variants to change over space and time. In summary, my research is focused on developing statistical models for proportions of genetic variants over space and time in the presence of evolution. SARS-CoV-2 will be used to motivate the development of the methods, but the methods themselves are the object of study. This research will contribute to the body of statistical methodology for proportion data, will provide a statistical summary of the COVID-19 pandemic, and produce tools for wastewater surveillance of other viruses and/or bacteria.