Kernel-DMD for multiome data integration and control

Abstract

Research in multiome data integration comes with the challenge of high-dimensionality and a small sample size in time series data. Traditional statistical tools often fail to capture true functional modules in large molecular networks, resulting in spurious associations. Dynamical systems theory overcomes this hurdle by assum ing the biological system follows a trajectory that can be modelled in such a way that the interactions in the network have a causal nature and pertain to mechanistic processes. Here we use kernel-DMD, a data-driven dynamical systems tool for time series data, for multiome network integration in the exotic plant species Clusia. We uncover differing modes of photosynthesis that correspond to the C3-like or strong CAM dynamics of two species, Clusia major and Clusia rosea and implement a control strategy that enables the in silico phenocopying between the two species. We demonstrate the applicability of the Koopman operator to multiome data integration, uncover drivers of plasticity in molecular networks and also identify key biomarkers that could potentially establish more resilient forms of photosynthesis, such as CAM, for the introduction of new crop bioengineering possibilities in C3 plants.