Bioinformatics Seminar
Time: 11AM
Venue: Davis Auditorium and Online
9 April 2024
LUPINE (LongitUdinal modelling with Pls for NEtwork inference)
Saritha KodikaraMIG/The University of Melbourne
Our objective was to infer taxa associations in a longitudinal setting, which is important for understanding the coexistence, competition, collaboration, and influence among different taxa. This understanding also allows us to observe how targeted interventions, such as dietary changes and antibiotic usage, alter these associations. However, traditional metrics such as correlation fall short due to unique data characteristics, including its compositional nature, high dimensionality, and sparsity. To address these challenges, we propose LUPINE, a longitudinal network inference method that leverages conditional independence and low-dimensional data representation. LUPINE employs a sequential approach to generate networks while considering information from all time points. These networks highlight key taxa that vary between treatment groups and time points. We validate LUPINE using simulated data and applied it to three case studies, yielding biologically meaningful insights.