A0993
Title: A Bayesian functional factor model for high-dimensional molecular curves
Authors: Salima Jaoua - University of Zurich (Switzerland)
Helene Ruffieux - Univerisity of Cambridge (United Kingdom) [presenting]
Daniel Temko - University of Cambridge (United Kingdom)
Abstract: The increasing availability of longitudinal measurements on gene products is set to improve the understanding of the molecular processes underlying disease risk and progression. However, methods for modelling coordinated patterns of temporal variation are currently lacking. A Bayesian approach for representing high-dimensional curves is proposed, combining latent factor modelling and functional principal component analysis (FPCA). This approach captures correlations across variables (gene products) and time by positing that a subset of variables contributes to a small number of FPCA expansions (representing latent disease processes) through variable-specific loadings. Subject variability is modelled using a few functional principal components, each characterised by a smoothly varying temporal function. A variational inference algorithm is introduced, with analytical updates, coupling efficiency and principled parameter uncertainty quantification. Extensive numerical experiments illustrate the ability of the approach to (i) accurately estimate variable-specific loadings, FPCA latent functions and subject-specific component scores and (ii) scale to realistic molecular data sizes. This framework should help disentangle disease heterogeneity by clarifying how gene pathways coordinate over time and predicting molecular trajectories at the subject level, thus facilitating targeted interventions and personalized treatments.
