UCLA

A suite of methods known as tensor decompositions have enabled researchers to explore high-order biological interactions more effectively. These methods are essential for allowing data-analysts to keep up with the rapid generation of high-throughput data the scientific community is now capable of producing. However, few tensor decomposition methods are capable of handling longitudinal data, one of the most common data types in biology. Here, we have developed a tensor factorization method capable of factoring data with a continuous mode, to combat the issue of reducing data with measurements taken over time. The method is capable of capturing a large amount of variance in both simulated and published longitudinal immune response data. The method additionally reproduces specific temporal dynamics in the data with high accuracy. Finally, it allows us to better visualize the key biological patterns and their interactions with time across the entire data-set.