A1276
Title: Bayesian nonparametric priors and discrete random structures for high-dimensional biological inference
Authors: Francesco Denti - University of Padova (Italy) [presenting]
Abstract: Discrete random structures within Bayesian nonparametric frameworks are employed to address inference challenges posed by two complex, state-of-the-art biological imaging datasets. The first application focuses on MALDI-mass spectrometry imaging. By modeling nested, separately exchangeable data, a discrete biclustering structure is induced that simultaneously partitions spatial locations with similar abundance profiles and molecules with similar expression patterns. To ensure these discrete clusters correspond to biologically meaningful regions, a hidden Markov random field prior is incorporated. This structural regularization facilitates the precise detection of molecular activation patterns and yields an interpretable spatial segmentation. The second model addresses neural activity recorded via calcium imaging in freely moving animals. A spatiotemporal mixture model uncovers discrete networks of co-activating neurons over time. This is achieved through a multivariate time-series framework that detects calcium trace spikes and captures recurring temporal patterns. To inform the clustering structure with network topology, a Dependent Dirichlet Process is employed that embeds anatomical proximity between neurons. Concurrently, a standard Dirichlet Process is used to segment spike amplitudes into discrete latent states, partitioning neural activity by magnitude.
