Uncovering hidden biological processes by probabilistic filtering of single-cell data

Elucidating underlying biological processes in single-cell data is an ongoing challenge and the number of methods that recapitulate dominant signals in such data has increased significantly. However, cellular populations encode multiple biological attributes, related to their spatial configuration, temporal trajectories, cell-cell interactions, and responses to environmental cues, which may be overshadowed by the dominant signal and thus much harder to recover. To approach this task, we developed SiFT (SIgnal FilTering), a method for filtering biological signals in single-cell data, thus uncovering underlying processes of interest. Utilizing existing prior knowledge and reconstruction tools for a specific biological signal, such as spatial structure, SiFT filters the signal and uncovers additional biological attributes. SiFT is applicable to a wide range of tasks, from the removal of unwanted variation in the data as a pre-processing step to revealing hidden biological structures. Applied for pre-processing, SiFT outperforms state-of-the-art methods for the removal of nuisance signals and cell cycle effects. To recover underlying biological structure, we use existing prior knowledge regarding liver zonation to filter the spatial signal from single-cell liver data thereby enhancing the temporal circadian signal the cells are encoding. Lastly, we showcase the applicability of SiFT in the case-control setting for studying COVID-19 disease. Filtering the healthy signal, based on reference samples from healthy donors, exposes disease-related dynamics in COVID-19 data and highlights disease informative cells and their underlying disease response pathways.